ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022

using Low Cost Devices (Dec 5‐16, 2022)

BIOMETRIC RECOGNITION
CHALLENGES
AND
OPPORTUNITIES

Dr. Surya Prakash

Associate Professor
Department of Computer Science & Engineering
Indian Institute of Technology Indore
[email protected]

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 1
Cost Devices (Dec 5-16, 2022)

Introduction
 Biometrics:
– It refers to metrics related to human characteristics and traits.
– Used to recognize/distinguish a person from another
 Examples:
– Face, fingerprint, iris, hand geometry, Ear, palm print, retina, DNA etc.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 2
Cost Devices (Dec 5-16, 2022)

1
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Introduction – Various Traits

 Various traits: Two types
– Physiological:
• These characteristics are related to the shape of the body.
• Examples: fingerprint, face recognition, DNA, palm print, hand
geometry, iris recognition, retina and odor/scent.
– Behavioral:
• These characteristics are related to the pattern of behavior
of a person.
• Example typing rhythm, signature, gait, and voice.
• Some researchers have coined the term behaviometrics to
describe the latter class of biometrics.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 3
Cost Devices (Dec 5-16, 2022)

Introduction - what is Biometric Recognition?

 Automated recognition of
individuals based on their
biometric characteristics
 Physiological (biological)
 Behavioral
 Biometric features extracted from
the biometrics traits
 Distinguishable
 Repeatable

H.T. F. Rhodes, Alphonse Bertillon: Father of Scientific Detection, Harrap, 1956

ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 4
Cost Devices (Dec 5-16, 2022)

2
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Ideal Biometric Trait

Uniqueness:
 The trait should be sufficiently different for individuals in the relevant
population such that they can be distinguished from one another.

Permanence:
 The manner in which a trait varies over time.
 More specifically, a trait with 'good' permanence will be reasonably invariant
over time with respect to the specific matching algorithm.

Measurability :
 The ease of acquisition or measurement of the trait.
 In addition, acquired data should be in a form that permits subsequent
processing and extraction of the relevant feature sets.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 5
Cost Devices (Dec 5-16, 2022)

Ideal Biometric Trait …contd

Performance:
 The accuracy, speed, and robustness of technology used.

Acceptability:
 How well individuals in the relevant population accept the technology such that
they are willing to have their biometric trait captured and assessed.

Circumvention:
 The ease with which a trait might be imitated using an artifact or substitute.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 6
Cost Devices (Dec 5-16, 2022)

3
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Comparison of Different Biometric Traits

Fingerprint Face Iris Voice Finger Vein
Uniqueness High Medium High Low High
Permanence Medium High Medium Low High
Measurability
Medium High Medium Medium Medium
(Collectability)
Performance Medium Low High Low High
Speed and cost High Low High Low Medium
Acceptability Medium High Low High High
Circumvention Medium Low Low High Low

Long term Stability Low Low Medium Low High

Size of template Small Large Small Small Small

ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 7
Cost Devices (Dec 5-16, 2022)

Comparison of Different Biometric Traits

Image source: Internet

ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 8
Cost Devices (Dec 5-16, 2022)

4
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Comparison of Different Biometric Traits

 Cost vs. Accuracy

Iris
Retina
Signature
Cost

Face Fingerprint

Voice

Accuracy

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 9
Cost Devices (Dec 5-16, 2022)

Comparison of Different Biometric Traits

 Obtrusive vs. Non-obtrusive

Image source: Internet

ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 10
Cost Devices (Dec 5-16, 2022)

5
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Why do we need biometrics?

 Traditionally methods for Recognition: Possession & token-based
– Possession based: (that we have) ID card, smart card, credit card
– Token based: (that we know) Password, PIN
– Issues:
• People forget ID and Password.
• Its easy to crack Password.

 Alternative: Biometric based Recognition

– Ensure uniqueness.
– We don’t need to remember our biometrics
– Difficult to imitate
– Physical presence of the person required for authentication

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 11
Cost Devices (Dec 5-16, 2022)

Working of Biometric System

 There are two steps involved in biometric authentication:
– Enrollment:
• the biometric characteristic and/or its related information is extracted and
stored in the database as a template.
– Authentication (Verification/Identification):
• the biometrics in query are compared with the existing template to measure
the degree of similarity.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 12
Cost Devices (Dec 5-16, 2022)

6
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Verification vs. Identification

 Verification: Am I who I claim to be?
– One-to-one comparison
– Verification outcomes
• Yes/No or True/False
• Can confirm or deny the specific identification claim of a person

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 13
Cost Devices (Dec 5-16, 2022)

Verification vs. Identification

 Identification: Who am I?
– One-to-many comparison
– Identification Outcomes:
• can determine the identity of a person from a biometric database without
that person first claiming an identity.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 14
Cost Devices (Dec 5-16, 2022)

7
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Verification vs. Identification

Verifying

vs.

Determining

the Identity

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 15
Cost Devices (Dec 5-16, 2022)

Identifiable Biometric Data vs. Biometric Templates

 Depending on when they are generated, templates can be

referred to as enrollment templates or match templates.
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 16
Cost Devices (Dec 5-16, 2022)

8
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Components of a Biometric System

Enrollment
Database

Feature Template Verification

Preprocessing Matcher
Extraction Generation

Accept/Reject
Sensor

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 17
Cost Devices (Dec 5-16, 2022)

Components of a Biometric System

 Sensor:
 To acquire biometric data

 Feature extractor:
 To extract a set of discriminative features from the data

 Matcher:
 To compare two extracted feature sets

 Database:
 To store biometric templates of individuals

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 18
Cost Devices (Dec 5-16, 2022)

9
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Identity vs. Recognition

 We donot necessarily want to get identity
 We want to recognize a person

???
Based on a single fingerprint
image, we cannot say this belongs
INPUT to person XYZ

We need a reference fingerprint

XYZ image that is known to belong to
REFERENCE person XYZ in order to make this
assessment

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 19
Cost Devices (Dec 5-16, 2022)

Multimodal Biometric System

A prototype multimodal biometric system

Image source: Internet
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 20
Cost Devices (Dec 5-16, 2022)

10
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Fusion in Biometrics
 Fusion at the feature level:

– Each system provides a set of features (a feature vector).

– These features are concatenated, and combined feature set

(concatenated feature vector) is used to assert the veracity of the
claimed identity.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 21
Cost Devices (Dec 5-16, 2022)

Fusion in Biometrics
 Fusion at the matching score level:

– Each system provides a matching score indicating the proximity of

the feature vector with the template vector.

– These scores can be combined to assert the veracity of the claimed

identity.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 22
Cost Devices (Dec 5-16, 2022)

11
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Fusion in Biometrics
 Fusion at the decision level:

– Each sensor can capture multiple biometric data and the resulting
feature vectors individually classified into the two classes –– accept
or reject.

– A majority vote scheme can be used to make the final decision.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 23
Cost Devices (Dec 5-16, 2022)

Fusion in Biometrics
 Fusion in the context of biometrics can take the following
forms :

– (1) Single biometric multiple representation

– (2) Single biometric multiple matchers

– (3) Multiple biometric fusion

Details follows…..

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 24
Cost Devices (Dec 5-16, 2022)

12
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Fusion in Biometrics
 (1) Single biometric multiple representations

– This type of fusion involves using multiple representations on a

single biometric indicator.

– Typically, each representation has its own classifier.

– Example: one fingerprint, different types of features, for different

feature-based representation, different matcher

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 25
Cost Devices (Dec 5-16, 2022)

Fusion in Biometrics
 (2) Single biometric multiple matchers

– It is also possible to incorporate multiple matching strategies in the

matching module of a biometric system and combine the scores
generated by these strategies.
– Example: one fingerprint, one feature set, same feature set is used in
different classifiers.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 26
Cost Devices (Dec 5-16, 2022)

13
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Fusion in Biometrics
 (3) Multiple biometric fusion

– By integrating matching scores obtained from multiple biometric

sources.

– These include majority voting, sum and product rules, k-NN

classifiers, SVMs, decision trees, Bayesian methods, etc.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 27
Cost Devices (Dec 5-16, 2022)

Fusion in Biometrics
 (4) Others

– 1. Store multiple templates in database.

• Example : A fingerprint biometric system may store multiple templates of a

users fingerprint (same finger) in its database.
• When a fingerprint impression is presented to the system for verification, it
is compared against each of the templates, and the matching score generated
by these multiple matchings are integrated.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 28
Cost Devices (Dec 5-16, 2022)

14
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Fusion in Biometrics
 (4) Others

– 2. A system may store a single template of a users

finger, but acquire multiple impressions of the
finger during verification.

– 3. Another possibility would be to acquire and use

impressions of multiple fingers for every user.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 29
Cost Devices (Dec 5-16, 2022)

Performance Evaluation of Biometrics Systems

 Different Metrics
– False Acceptance Rate (FAR)
– False Rejection Rate (FRR)
– Equal Error Rate (EER)
– Failure-to-Enroll rate
– Receiver Operating Characteristics (ROC) Curve

 No single metric indicates how well a biometric system or device

performs.

 Analysis of different metrics is necessary to assess the

performance of a specific technology.
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 30
Cost Devices (Dec 5-16, 2022)

15
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

False Acceptance Rate

 If Rahul enters Ramesh’s username or ID, presents his
biometric data, and successfully matching as Jane Doe.
– This is classified as false acceptance.

 The probability of this happening is referred to as false

acceptance rate (FAR)[ stated as percentage]

 This is because two people have similar enough biometric

characteristics – a fingerprint, a voice, or a face – that the system
finds a high degree of correlation between the users’ template.
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 31
Cost Devices (Dec 5-16, 2022)

False Acceptance Rate

 FAR can be reduced by adjusting the thresholds but the false
rejection rate will increase.

 Example:
– A system with a false acceptance rate of 0 percent, but false rejection rate
of 50 percent, is secure but unusable.

 False acceptance rate is the most critical accuracy metric because

an imposter break-in will certainly be a more attention-getting
event than other failings of a biometric system.

 The most important false match metric in real-world deployments

is the system false match rate.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 32
Cost Devices (Dec 5-16, 2022)

16
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

False Rejection Rate

 If Rahul enters his username or ID, presents his biometric data to
a biometric system, and fails to match.
– This is classified as false rejection.
–
 The probability of this happening is the false rejection rate (FRR).

 This can be attributed to changes in user’s biometric data, changes

in how a user presents biometric data, and changes in the
environment in which data is presented.

 High FRR will result in lost productivity, frustrated users, and an

increased burden on help desk or support personnel.
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 33
Cost Devices (Dec 5-16, 2022)

Reasons of FRR
 Changes in user’s biometric data
– Voice-scan system is influenced by sore throats
– Facial-scan system is affected by changes in light
– Fingerprint changes over time, scars, aging and general
wear.

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 34
Cost Devices (Dec 5-16, 2022)

17
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Acceptance and Rejections – Summary

 If someone else is trying to verify as you, the system
would try to match the two templates.
– If the two templates were to match – this is classified as
false acceptance.
– If your authentication template fails to match your enrolled
template, then this is referred to as a false rejection.
– If you are new and fail to enroll to a biometric system, this
is called – failure to enroll (FTE).

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 35
Cost Devices (Dec 5-16, 2022)

Accuracy Rates
 Single False Acceptance Rate vs. System False
Acceptance Rate
– If the FAR is 1/10,000 but you have 10,000 templates on
file — odds of a match are very high

 Ability to Verify (ATV) rate:

– % of user population that can be verified
– ATV = (1-FTE)(1-FRR)

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 36
Cost Devices (Dec 5-16, 2022)

18
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Receiver Operating Characteristic (ROC) Curve

 Cost/benefit analysis
of decision making.

True acceptance rate

 Tradeoff between true
acceptance rate and Legitimate
users get
false rejection rate. accepted.

Legitimate users False rejection rate

get rejected.
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 37
Cost Devices (Dec 5-16, 2022)

Privacy and Biometrics

 What else does your Biometric Data reveal?
 Altering Biometric Data to enhance privacy and retain
utility

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 38
Cost Devices (Dec 5-16, 2022)

19
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Biometric Fusion
 Biometrics + Demographics + Spoof Detector +
Quality
 Primary Biometrics + Soft Biometrics

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 39
Cost Devices (Dec 5-16, 2022)

Challenges in Biometrics: Robust Matching

 Robust matching algorithms

Same Person Different People Same or Different?

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 40
Cost Devices (Dec 5-16, 2022)

20
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Challenges in Biometrics: Attacks

 Robust to presentation attacks
– Ensuring that the input data is uncorrupted and from a real
person

Real Fingerprint

EcoFlex Gelatin Latex Silgum WoodGlue

Spoof Fingerprints

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 41
Cost Devices (Dec 5-16, 2022)

Presentation Attacks
Modality Presentation Attacks
Face Printed Photo Attack Low
Replay Attack Low
3D Mask Attack High
Facial Cosmetics Attack Low (Obfuscation)
High (Spoofing)
Plastic Surgery High
Fingerprint Gummy Finger Attack Printed Medium
2D Fingerprints Medium
3D Printed Targets High
Fingerprint Tape Medium
Cadaver Fingers High
Altered Fingerprints High
Iris Printed Photo Attack Video Medium
Replay Attack Medium
Textured Contact Lenses Medium
Prosthetic Eyeballs High
Iris Surgery High
Drug-induced Pupil Dilation Medium

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 42
Cost Devices (Dec 5-16, 2022)

21
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Presentation Attack Detection

© Dr. Xiaoming Liu, Michigan State University | http://cvlab.cse.msu.edu/project-face-anti.html

ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 43
Cost Devices (Dec 5-16, 2022)

Challenges in Biometrics: Security and Privacy

 Robust Security and Privacy

Othman, A., Ross, A. (2015). Privacy of Facial Soft Biometrics: Suppressing Gender But Retaining Identity. In:
Proceedings of ECCV 2014 Workshops (ECCV 2014), LNCS vol 8926. Springer https://doi.org/10.1007/978-3-319-16181-5_52
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 44
Cost Devices (Dec 5-16, 2022)

22
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Challenges in Biometrics: Security of Biometric Data

 How can biometric templates be stored and
transmitted securely?
 Protecting the biometric templates in the database
 Biometric Template Security

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 45
Cost Devices (Dec 5-16, 2022)

Challenges in Biometrics: Privacy of Biometric Data

 Ensuring the privacy of an
individual
 Age, Gender, Ethnicity, can be
automatically derived from the
face image
• Gender: Male
 That is, a trained classifier or a
• Age: 25
regressor may be used to • Health: Very good
automatically deduce certain soft • Eye Sight: Wears glasses
• Ethnicity: Asian Indian
biometric attributes
For details, see Dantcheva, Elias, Ross, “"What Else Does Your Biometric Data Reveal? A Survey on Soft Biometrics,” TIFS 2016
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 46
Cost Devices (Dec 5-16, 2022)

23
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Challenges in Biometrics: Noise

 Robust to Sensor Size
– What types of signal enhancement and matching models are
necessary to conduct biometric recognition using severely degraded
biometric data?

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 47
Cost Devices (Dec 5-16, 2022)

Challenges in Biometrics: Contrast, Illumination

 3D Biometrics
– 3D data contains (x, y, z) values (can have
color/texture as well)

– Insensitive to imaging problems such as

lighting and shadows

– Can provide geometric information of the

object

– 3D data can handle more general pose

variations
Image source: http://cool-3d-pictures.blogspot.com/2010/10/3d-face.html
https://in.pinterest.com/pin/519321400778202577/
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 48
Cost Devices (Dec 5-16, 2022)

24
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Challenges in Biometrics: Scale

 Robust to Scale
– Aadhaar: Stated to be the world's
largest biometric ID system
– ~1.27 billion enrolled members as of
23 December 2020
 How can large biometric
databases be efficiently searched?
– Required to rapidly locate an
identity of interest
– Indexing of Biometric Databases
Image source: https://thelogicalindian.com/news/government-approves-aadhar- bill-the-benefits-risk-of-aadhar/
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 49
Cost Devices (Dec 5-16, 2022)

Challenges in Biometrics: Partial Fingerprints

 Small sensors
 Capture a limited portion of full finger
 Multiple partial
 fingerprints are captured
 Enroll multiple fingers
 Access granted
 If the sensed partial fingerprint matches any one of the partial
fingerprint of any enrolled finger

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 50
Cost Devices (Dec 5-16, 2022)

25
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Challenges in Biometrics: MasterPrints

 Fingerprints that by chance (rather than intention) match
with a large proportion of the fingerprint population
 Could be either full prints or partial prints

Roy, Memon, Ross, “MasterPrint: Exploring the Vulnerability of Partial Fingerprint-based Authentication Systems,” TIFS 2017
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 51
Cost Devices (Dec 5-16, 2022)

Challenges in Biometrics: MasterPrints

 With a dictionary of 5 MasterPrints, it was possible to attack
– 26.46% users (each having 12 impressions per finger) in the
FingerPass DB7 capacitive fingerprint dataset
– 65.20% users (each having ≈ 80 partial impressions per finger) in
the FVC optical fingerprint at an FMR of 0.1%.

 The attack accuracy varied greatly with the FMR value and
the number of impressions per finger

Roy, Memon, Ross, “MasterPrint: Exploring the Vulnerability of Partial Fingerprint- based Authentication Systems,” TIFS 2017
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 52
Cost Devices (Dec 5-16, 2022)

26
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Other Opportunities: Smartphone - Continuous Authentication

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 53
Cost Devices (Dec 5-16, 2022)

New Opportunities: Identification Without Biometric Data

 With just anonymous location data, it is possible to figure out “who you are” by
tracking your smartphone
 15 months of mobility data for 1.5 million individuals and found that human
mobility traces are highly unique.
 4 spatio-temporal points are enough to uniquely identify 95% of the individuals

De Montjoye, Hidalgo, Verleysen & Blondel, “Unique in the Crowd: The Privacy Bounds of Human Mobility”, Scientific Reports, vol. 3, 2013
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 54
Cost Devices (Dec 5-16, 2022)

27
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

New Opportunities: From Image to Sensor

IMAGE SENSORS

Unit Level versus Brand Level

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 55
Cost Devices (Dec 5-16, 2022)

New Opportunities: Dual Factor Authentication

Device ID + Subject ID

Who you
are?

What
device you
have?

Arjona et al., "Securing Minutia Cylinder Codes for Fingerprints through Physically Unclonable Functions: An Exploratory Study," ICB 2018
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 56
Cost Devices (Dec 5-16, 2022)

28
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Privacy - Decomposing a Binary Image

 Decomposing a fingerprint into two random images using
Visual Cryptography

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 57
Cost Devices (Dec 5-16, 2022)

Privacy - Face Visual Cryptography

Actual Face HOST

IMAGE IN
SERVER 1

Simple XOR operator

HOST
IMAGE IN
SERVER 2

Ross and Othman, “Visual Cryptography for Biometrics Privacy”, TIFS 2011
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 58
Cost Devices (Dec 5-16, 2022)

29
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Privacy - Face Morphing

 To generate a mixed face image, the principle of face morphing
is used
 The mixed face image can be anywhere along a continuum
from F1 to F2

F1
MF

F2

Othman and Ross, “Privacy of Facial Soft Biometrics: Suppressing Gender But Retaining Identity”, ECCV Workshop, 2014
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 59
Cost Devices (Dec 5-16, 2022)

Soft Biometric Privacy

 Gender attribute of an input face image is progressively
suppressed
 With respect to a face matcher the identity is preserved

Name Alice Alice Alice Alice

Gender Female Female Male Male
(confident) (less confident) (less confident) (confident)

Othman and Ross, “Privacy of Facial Soft Biometrics: Suppressing Gender But Retaining Identity”, ECCV Workshop, 2014
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 60
Cost Devices (Dec 5-16, 2022)

30
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

SAN: Semi-Adversarial Network

Mirjalili et al., “Semi-Adversarial Networks: Convolutional Autoencoders for Imparting Privacy to Face Images,” ICB 2018
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 61
Cost Devices (Dec 5-16, 2022)

SAN: Semi-Adversarial Network

Mirjalili et al., “Semi-Adversarial Networks: Convolutional Autoencoders for Imparting Privacy to Face Images,” ICB 2018
ATAL FDP on Research Avenues in Biometrics using Low
Dec 5, 2022 62
Cost Devices (Dec 5-16, 2022)

31
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

References
 Othman, A., Ross, A. (2015). Privacy of Facial Soft Biometrics: Suppressing Gender But
Retaining Identity. In: Proceedings of ECCV 2014 Workshops (ECCV 2014), LNCS vol 8926.
Springer https://doi.org/10.1007/978-3-319-16181-5_52
 Othman and Ross, “Privacy of Facial Soft Biometrics: Suppressing Gender But Retaining Identity”,
ECCV Workshop, 2014
 Mirjalili et al., “Semi-Adversarial Networks: Convolutional Autoencoders for Imparting Privacy to
Face Images,” ICB 2018
 Ross and Othman, “Visual Cryptography for Biometrics Privacy”, TIFS 2011
 Arjona et al., "Securing Minutia Cylinder Codes for Fingerprints through Physically Unclonable
Functions: An Exploratory Study," ICB 2018
 Arjona et al., "Securing Minutia Cylinder Codes for Fingerprints through Physically Unclonable
Functions: An Exploratory Study," ICB 2018
 De Montjoye, Hidalgo, Verleysen & Blondel, “Unique in the Crowd: The Privacy Bounds of Human
Mobility”, Scientific Reports, vol. 3, 2013
 Roy, Memon, Ross, “MasterPrint: Exploring the Vulnerability of Partial Fingerprint- based
Authentication Systems,” TIFS 2017
 https://www.biometricsinstitute.org/wp-content/uploads/RossCanberra_Feb2018.pdf
 https://www.biometricsinstitute.org/

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 63
Cost Devices (Dec 5-16, 2022)

Thank you!

ATAL FDP on Research Avenues in Biometrics using Low

Dec 5, 2022 64
Cost Devices (Dec 5-16, 2022)

32
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Ear as a Modality: Path and Progress

(3D Ear Based Human Recognition)

Dr. Surya Prakash

Department of Computer Science & Engineering
Indian Institute of Technology Indore

Outline
• Introduction
• Issues with 2D ear recognition
• Motivation
• Techniques for ear recognition
– 3D ear recognition using global and local features
– Ear recognition in 3D using 2D curvilinear features
– Geometric statistics based descriptor for 3D ear
recognition
– Age invariant analysis of human ear
• Conclusion
• Future Directions

1
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Ear as a biometric
• The acquisition procedure is contactless and nonintrusive.

• The ear is robust to age, beautifiers, and outward appearances, while

other biometrics may be affected by the mentioned variations.

• Another appealing property of ear biometrics is its distinctiveness [1]

(example: features of the ear are distinct for identical twins [2]).

• Ear images can also serve as supplements for other biometric modalities
(example: face recognition technology may struggle with profile faces, the
ear can serve as a source of information).

• These all facts has significant implications for security related applications
and put ear images at par with other biometric modalities.
[1] A. Jain, A. Ross, K. Nandakumar, Introduction to biometrics, Springer Science & Business Media, 2011.
[2] H. Nejati, L. Zhang, T. Sim, E. Martinez‐Marroquin, G. Dong, Wonder Ears: Identification of Identical Twins from Ear Images, in:
Proceedings of the International Conference on Pattern Recognition, IEEE, 2012, pp. 1201–1204.

Ear Recognition in 2D and 3D

• Use of 2D images for recognition

• Use of 3D images for recognition

2
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Why 3D Ear?
Problems with 2D ear images
Illumination

Pose variations
Pose variations can be addressed with suitable alignment techniques. 3D data and generic
3D models are exploited to normalize the geometry of the ear.

Scaling
12/12/2022 PAMIL

3D Ear Recognition
• Mostly based on 3D image registration

3
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Rigid vs. Non‐rigid

Distances between all points
remain constant.

Rigid shapes Non-rigid shape

(Main app: 3D scan registration) vs. (Main apps: shape completion &
attribute transfer via shape
correspondence)
Ref: Crum, W.R., Hartkens, T. and Hill, D.L.G.. Non‐rigid image registration: theory and practice. The British journal of radiology, 77
(suppl_2), pp. 140‐153, 2004.

Rigid Alignment

If correct correspondences are known, Shape 1

(Data)
it is possible to find correct relative
rotation/translation p2

p1 p3 p4 p5

q5

q4
q2 q3
q1

Shape 2
(Model)

Problem Formulation:
Given two sets points: in . Find the rigid transform:
and that minimizes:

4
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Correspondence and Matching

Alignment
• Method 1: Principal component analysis (PCA)
– Aligning principal directions

• Method 2: Singular value decomposition (SVD)

– Optimal alignment given prior knowledge of
correspondence

• Method 3: Iterative closest point (ICP)

– An iterative SVD algorithm that computes
correspondences as it goes

5
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Method 1: PCA
• Compute a shape‐aware coordinate system for each model
– Origin: Centroid of all points
– Axes: Directions in which the model varies most or least
• Transform the source to align its origin/axes with the target

Slide Courtesy: https://www.cse.wustl.edu/~taoju/cse554/lectures/lect05_Alignment.ppt

Method 1: PCA
• Limitations
– Centroid and axes are affected by noise
Noise

Axes are affected PCA result

6
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Method 1: PCA
• Limitations
– Axes can be unreliable for circular objects
• Eigenvalues become similar, and eigenvectors become
unstable

Rotation by a small angle PCA result

Method 2: SVD
• Optimal alignment between corresponding
points
– Assuming that for each source point, we know
where the corresponding target point is

7
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

Method 3: ICP
• The idea
– Use PCA alignment to obtain initial guess of
correspondences
– Iteratively improve the correspondences after
repeated SVD
• Iterative closest point (ICP)
– 1. Transform the source by PCA‐based alignment
– 2. For each transformed source point, assign the
closest target point as its corresponding point. Align
source and target by SVD.
• Not all target points need to be used
– 3. Repeat step (2) until a termination criteria is met.

Ref: Besl and McKay, “A method for registration of 3D shapes”, IEEE Transaction on Pattern Analysis and Machine
Intelligence, vol.11,No.2, May 1992.

16

Data

Model

8
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

17

Pick points on data surface

18

Find closest points on stationary surface

9
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

19

Measure the total distance

20

Register point sets (rigid)

10
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

21

Register point sets (rigid)

22

Find (new) closest points

11
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

23

Find (new) closest points

24

Register Points

12
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

25

Register Points, and so on…

26

Different representations of 3D Ear

Point cloud Mesh

Mesh Voxel
12/12/2022

13
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

27

Database used in evaluation

Sample 3D ear images from UND‐J2 database

28

IIT Indore database

2D Data collection:
• Images collected from students , faculty and staff at Indian Institute of Technology
Indore; and students from government and private schools around Indore.
• 2D and 3D Data collected over a three‐year period.
• The images were taken in an indoor environment.
• The database has been acquired from different subjects, all of them aged 5‐60
years.
• Four right ear images, four left ear images and two full profile images have been
taken for each individual.
• All images were taken with Sony DSC‐WX350 and Nikon 7200.

Sample 2D ear images from IIT Indore database

Acknowledgement of data collection to my colleagues Mr. Akhilesh, Mr. Uttam, Dr. Sadaf and Mr. Ishan

14
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

29
Equipment and experimental setup used in
3D scan

Artec Eva
3D Scanner

12/12/2022 30

2D face Images

Sample 2D face images from IIT Indore

15
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

31

3D Data collection
Images were taken of students, teachers and staff from the
Indian Technology Institute Indore, Indore, India.
The images have been obtained in an indoor environment.

Samples of 3D face images from IIT Indore

Samples of 3D ear images from IIT Indore

12/12/2022

32

Comparison of the existing ear database (2D)

Database Year Subject Images Sides Occlusion Gender Ethnicity
CP 1995 17 102 left none Both White

USTB I 2002 60 185 right none Both Asian

USTB II 2004 77 308 right none Both Asian

IITD I 2007 125 493 right none Both Asian

AMI ‐ 100 700 both mild Both White

WPUT 2010 501 2071 both severe Both White

IITD II ‐ 221 793 right none Both Asian

AWE 2016 100 1000 both severe Both Various

IIT Indore 2016 659 5272 both mild Both Asian

12/12/2022 PAMIL

16
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

33

Comparison of the existing ear database (3D)

No. Referenc Data Available Acquisition # Subjects # Images Average
e Device no. of
Vertices
1 Yan and Range Images Minolta 415 1800 ~ 6000
Bowyer (480x640), Vivid 910
Co‐registered
2D
2 IIT Point cloud, Artec‐Eva Phase1 – 188 1128 ~ 8000
Indore Mesh Phase2 – 176 1056
Phase3 – 188 1128

3 Chen & Range Images Minolta 155 902 ‐

Bhanu (200x200), vivid 300
Co‐registered
2D
4 Yahui Liu Range Imgaes Custom 250 2000 ‐
(140x200) device

34

IIT Indore database statistics

Face
Ear

17
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

35

Motivation for descriptor based matching

Given an example:

Find the most

similar ear …
in a database , ,

36

Descriptor based problem definition

• We need a Descriptor

D: → D( )
• We need a Distance Measure :
– Close to (application driven) notion of resemblance
– Computationally cheap and robust

18
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

37

General outline of the descriptor based techniques

Feature Point Feature

3D Point Cloud
Detection Descriptor
(Model)
(Model) (Model)
Feature
Matching
(Score
Calculation)
Feature Point Feature
3D Point Cloud
Detection Descriptor
(Test)
(Test) (Test) score  threshold

Recognition
Success

12/12/2022

38

General outline of the descriptor based techniques

Feature Point Feature

3D Ear
Detection Descriptor
(Gallery)
(Gallery) (Gallery)

Feature
Matching

Feature Point Feature

3D Ear
Detection Descriptor
(Probe)
(Probe) (Probe)

Matching ICP Correspondence

Score Registration Points

12/12/2022

19
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

39

3D Models
• Fourth generation of multimedia following sound,
images and videos.
• Represents 3D objects using collection of points
connected by various geometric entities.
• Classified into two categories:

Solid Model Shell/Boundary Model

Image courtesy : Mark Pauly, ETH Zurich

40
3D Descriptors – Literature survey
Sr.No Name of the Technique Data Type Used Performance
1 Splash Mesh Robust to noise
2 Point Signature Mesh Not unique, sensitive
to mesh resolution

3 Spin image Mesh Most cited method,

with a set of variants,
sensitive to mesh
resolution

4 Point’s fingerprint Mesh Outperforms (2) and

(3)
5 Surface signature Mesh Outperforms (1), (2)
and (3)
6 3DSC / HSC Point cloud Outperforms (3)
7 NBS Mesh Outperforms (3)
8 3D Tensor Mesh Outperforms (3)
9 LSP Depth Image Comparable to (3)
12/12/2022

20
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

41

Sr.No Name of the Data Type Used Performance

Technique
10 THRIFT Point Cloud Sensitive to noise
11 Snapshot Mesh Outperforms (3)
12 VD‐LSD Point Cloud Generalized
descriptor
13 HMM Mesh Outperforms (3)
and (6)
14 Hua’s Mesh Suitable for surface
matching
15 EM Mesh Scale dependent
and invariant
descriptor
16 PFH Point cloud Time consuming
17 Spectral Feature Mesh Invariant to
isometric
deformation
18 2.5D SIFT Depth Image Outperforms SIFT
12/12/2022

42

Sr.No Name of the Data Type Used Performance

Technique
19 FPFH Point cloud More time efficient
than PFH
20 HKS Mesh Invariant to
isometric
deformation
21 MeshHOG Mesh

22 SHOT Mesh Outperforms (2),

(3) and (15)
23 USC Mesh Outperforms (6)

24 TriSI Mesh Outperforms (3), (9)

and (22)
25 RoPS Mesh Outperforms (3),
(9), (10), (21) and
(22)
12/12/2022

21
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

43

Ear Recognition Techniques

• Ear recognition using global and local features

• Ear Recognition 3D using 2D Curvilinear

Features

• Geometric statistics‐based descriptor for 3D

ear recognition

44

Ear recognition using global and local features

Global Local

Computing local descriptors in a dense A desirable characteristic of local

manner without detecting keypoints in an descriptor is the independent description
image results in encoding the global of keypoint locations, which makes partial
structure of the image, but at the expense matching of the extracted descriptors
of loosing robustness to partial occlusions possible and exhibit robustness to partial
(in most cases) occlusions and noise of the ear

Iyyakutti Iyappan G and Surya Prakash, 3D Ear Recognition using Global and Local Features, 7(3),
pp. 232-241, IET Biometrics, IET, 2018.

22
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

45

Ear recognition using global and local features

1) This technique is based on a combination of global and local descriptors.
2) The combined representation is used aiming at better results as compared to the
case when the local and the global descriptors are used in isolation.

(a) (b)
Histogram generation at point p (a) Three different
Sphere radii centered at point p. Red, blue, and
Input Data green points are in the annular regions between
the concentric spheres. (b) Histogram obtained
from the points in the annular regions.

46

Global feature descriptor

(a) (b)

(a) Eight neighboring spheres S1, S2, ……. S8 at a ring of radius R from the center sphere S0
located at point p

(b) Coded image generated using proposed LSGP (localsphere geom etry pattern)

12/12/2022 PAMIL

23
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

47

Final global descriptor vector

Computation of the proposed global descriptor vector. The coded image is partitioned
(red) into a number of blocks and each block is converted to a histogram. The final
feature vector (rightmost) is generated by concatenating all the histograms.

48

Identification rate on UND‐J2 database

12/12/2022 PAMIL

24
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

49

ROC for different descriptors

12/12/2022 PAMIL

50
Recognition system evaluation on UND‐J2
database

12/12/2022 PAMIL

25
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

51

Performance Comparison
Technique Database Gallery, Verification EER % Remarks
Probe accuracy %
Images
Hui and Chen UND ‐ F 302,302 ‐ 2.30 Small database,
Poor result
Syed MS UND – J2 415,415 94.00 4.10 Small database,
Islam et al. Poor result
Zhou Jindan UND – G 235,235 ‐ 1.60 Small database
et al.
Surya Prakash UND – J2 404,1376 98.30 1.80 Needs co‐registered
and Gupta et 2D ear image
al.

Ping Yan and UND – J2 415,1386 ‐ 1.20 Needs co‐registered

Kevin Bowyer 2D images,
concisely cropped
ear, Fails under pose
variations
12/12/2022 PAMIL

52

Performance Comparison cont..

Technique Database Gallery, Verification EER % Remarks
Probe accuracy %
Images
Xiaopeng Sun UND J2 415,1386 ‐ 4.00 No dependency on
et al. 2D data, however
poor result
Y Liu et al. In‐House ‐ ‐ 2.30 Poor result
database
Proposed UND J2 404,1376 98.69 1.56 Uses 3D data only
technique

12/12/2022 PAMIL

26
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

53

Ear Recognition 3D using 2D Curvilinear Features

1) This technique also based on keypoint detection and description

2) We use Co‐registered 2D & 3D ear images in recognition
3) Keypoints are detected using 2D images and recognition is performed using 3D

Iyyakutti Iyappan G, Surya Prakash, Syed Sadaf Ali, Piyush Joshi, Ishan R Dave and Akhilesh Mohan Srivastava, Ear Recognition in 3D
using 2D Curvilinear Features, 7(6), pp. 519-529, IET Biometrics, IET, 2018.

54

Co‐registered ear images

Sample of co‐registered 3D and 2D ear image from UND dataset: (a) 3D ear image, (b)
Co‐registered 2D ear image of (a), (c) 3D ear image of (a) in different poses for better
visualization

27
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

55

Steps in design process

1) Keypoint detection
– A one dimensional, extended polynomial filter is used to detect the
curvilinear ear structure.
– The pixels of curvilinear structure detected are further processed as
keypoints.
– The 2D image keypoints are mapped to the 3D image as keypoints

2) Keypoint description
a) A local descriptor is used to describe each keypoints.
b) The feature vectors are further used in matching ear pairs.

56

Filters used to obtain curvilinear features

12/12/2022

28
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

57

Filters applied with different orientation

12/12/2022

58

Curvilinear feature detection using filter

Curvilinear feature detection using filter F(θ, w) for different angles with constant
width: (a) θ = 0∘, w = 2, (b) θ = 45∘, w = 2 (c) θ = 90∘, w = 2, (d) θ = 135∘, w = 2, (e)
Final image with curvilinear structures computed using the maximum of the pixel
values at corresponding locations of all the four images (a, b, c, and d), (f) Binarised
image of curvilinear structures, (g) Thinned curvilinear structures superimposed on
original 2D ear image

29
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

59
Thinning and mapping of 2D keypoints onto 3D
ear

Mapping of 2D curvilinear features onto co‐registered 3D ear image: (a) Detected

curvilinear structures, (b) Thinned curvilinear structures, (c) Mapped key‐points (red) in 3D
image.

12/12/2022

60
Obtained Feature keypoints and correspondence
matching using the proposed descriptor

Feature keypoints highlighted Correspondence keypoints

in red color matching

12/12/2022

30
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

61

Matching of two 3D ear images

Demonstration of feature keypoints matching for the proposed technique. The co‐
registered 2D image along with the detected key‐points is side by the 3D image, where
the detected keypoints in 2D and the mapped keypoints in 3D are shown in blue.

62

PDF of matched and unmatched scores

Distribution of genuine and imposter scores on UND‐Collection J2

dataset

12/12/2022

31
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

63

Accuracy plots

FAR / GAR Threshold / Accuracy

12/12/2022

64
Sample Gaussian noise added ear images used
in experimentation

Curvilinear structure detection and mapping for noisy data: (a) Original data with
added Gaussian noise of varying from 0.1 to 0.5, (b) Detected curvilinear structures,
(c) Mapped curvilinear structure key‐points onto co‐registered 3D image.

12/12/2022

32
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

65
Samples of occluded ear images used in
the experimentation

Example of occluded ear images (a) Original 2D ear image (above) and the co‐
registered 3D image (below), (b‐e) shows 2D images (above) and corresponding 3D
images (below) for different amount of occlusion.

12/12/2022

66

Recognition on occluded and noisy ear images

Noisy images occluded images

12/12/2022

33
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

67

Performance Comparison
Technique Database Gallery, Verification EER % Remarks
Probe accuracy %
Images
Hui and Chen UND ‐ F 302,302 ‐ 2.30 Small database,
Poor result
Syed MS UND – J2 415,415 94.00 4.10 Small database,
Islam et al. Poor result
Zhou Jindan UND – G 235,235 ‐ 1.60 Small database
et al.
Surya UND – J2 404,1376 98.30 1.80 Needs co‐registered
Prakash and 2D ear image
Gupta et al.

Ping Yan and UND – J2 415,1386 ‐ 1.20 Needs co‐registered

Kevin Bowyer 2D images,
concisely cropped
ear, Fails under
pose variations
12/12/2022

68

Performance Comparison cont..

Technique Database Gallery, Verification EER % Remarks
Probe accuracy %
Images
Xiaopeng Sun UND J2 415,1386 ‐ 4.00 No dependency on
et al. 2D data, however
poor result
Y Liu et al. In‐House ‐ ‐ 2.30 Poor result
database
Proposed UND J2 404,1376 98.69 1.53 Uses 2D and 3D
technique data, uses loosely
cropped ear, no
effect of pose

12/12/2022

34
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

69
Geometric statistics‐based descriptor for 3D Ear
Recognition
1) Generalized 3D descriptors work well in discriminating
against different class of shapes.

2) However, generalized 3D descriptors do not perform

well if used in biometric applications to compare
extremely similar objects, such as 3D ear or face.

3) We propose an efficient feature keypoint detection and

description technique using geometric statistics to
represent 3D objects.

4) The effectiveness of the proposed technique is

demonstrated using 3D ears.
Iyyakutti Iyappan G and Surya Prakash, Geometric statistics-based descriptor for 3D ear
recognition, The Visual Computer Journal, pp. 1-13, Springer, 2018

70
Geometric statistics‐based descriptor for 3D Ear
Recognition
1) Keypoint detector
a) We find points in 3D ears with extreme variation in surface and these points are
called keypoints.
b) To find these points, we use a covariance matrix.
c) The covariance matrix is constructed at each point with the help of its neighbours,
from the support region of radius r.
d) Further, the matrix is decomposed to find Eigen values, λ0 ⩽ λ1 ⩽ λ2.

0
 t 0       (1 )
 0  1   2
2
 t1       ( 2 )
 0  1   2
t0  0 .0 1
t1  0 . 7

12/12/2022

35
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

71

Examples of keypoints for varying thresholds

72
Geometric statistics‐based descriptor for 3D Ear
Recognition
2) Keypoint descriptor
i) Place a plane perpendicular
to the normal of the
detected key point.
ii) Multi radii spheres centered
at keypoint with radii
r  meshresolution
iii) The geometric characteristics
of the neighbours falling in
the annular regions of the
multi radii spheres are
encoded as descriptor.
iv) There are three steps in
designing the descriptor.

12/12/2022

36
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

73
Projection of points from annular regions
to plane

74

How to encode the descriptor?

a) First step

Steps
a) Neighbours of a feature point projected to the plane,
b) 2D Histograms of the projected points in the annular regions of different sphere radius,
c) Lower order moments of 2D histograms,
d) Concatenation of the three histograms from (c)

12/12/2022
Ref: Mohammed Bennamoun, Ferdous A. Sohel, Yulan Guo, Min Lu, and Jianwei Wan. 2013. 3D free form object recognition using rotational projection statistics.
In Proceedings of the 2013 IEEE Workshop on Applications of Computer Vision (WACV) (WACV '13)

37
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

75

How to encode the descriptor?

b) Second step – use of normals

Steps
a) Slices with normals correspond to three different radius,
b) Histograms for each slice,
c) Concatenation of the three histograms from (b)

Ref: Zhou, Jindan, Steven Cadavid, and Mohamed Abdel‐Mottaleb. "A computationally efficient approach to 3d ear recognition
employing local and holistic features." CVPR 2011 WORKSHOPS. IEEE, 2011.

76

How to encode the descriptor?

c) Third step: use of signed distances

i) we use signed distances of the local neighbours lie in the multi radii of the feature
point to the plane. For spheres of multi radii (r * mesh‐resolution) the signed distances
are encoded separately as a histogram.

ii) All the histograms of the different radius are combined to form a descriptor.

12/12/2022

38
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

77

Experimental Results

98.60%
EER = 1.50%

FRR (False Rejection Rate): the proportion of genuine scores that are less than the threshold η.
FAR (False Acceptance Rate): the fraction of impostor scores that are greater than or equal to η.
GAR (Genuine Accepatance Rate): 1 ‐ FRR

12/12/2022

78

Conclusions
• 3D ear recognition has a potential to serve as a
good biometrics
• Robust to illumination and contrast
• Can be clubbed with other traits to get
superior recognition performance
• Invariant to aging:
• The analysis of age invariant nature of human ear has been carried
out using three phases of 2D and 3D ear data. It shows that ear is
robust to age variations up to quite extent.

39
ATAL FDP on Research Avenues in Biometrics 5‐Dec‐2022
using Low Cost Devices (Dec 5‐16, 2022)

79

Directions of Future Work

• Autonomous ear detection in 3D without using co‐

registered 2D images
• Age‐Invariant nature analysis of human ear
• Effect of inheritance in ear recognition
• Deep learning architecture for 3D ear recognition

80

Thank You

40
 Prospective Research
Proposals and Funding
Agencies

Dr. H. Honne Gowda

Former Special Secretary 9Technical)
Department of Science and Technology
Govt of Karnataka
Dec. 07, 2022
 Types of Grants
• Research Grants – Short, long-term, Specific Activity
• Seminar/Workshop Grants
• Training Program Grants
• Fellowship Grants
• Travel Grants
• Publication Grants
• Patent Grants
• Start-up Grants / Entrepreneurship
 Preparation Stage for Projects
• Preliminary Work: Thrust, Priority Areas & Contemporary Research
• Assess Your Own Bandwidth
• Literature Review, Discussion with Specialists/Experts
• Develop a specific, meaningful, actionable plan for what you
want to do and why you want to do it;
• Conducting Preliminary or Pilot Research
• Call for Applications by Various Funding Agencies (FA)
• Be Flexible with the Project Prioritization Process
• Understand the Guidelines, Eligibility, Terms and Conditions,
Application Procedures, before Applying
• Download all the Supporting Documents from their Website
• Process used by the Grantor Agency for Accepting
• Application before the Deadline
 Finding Prospective Grants
and Funding Agencies
• Whether your proposal receives funding will rely in large part on
whether your purpose and goals closely match the priorities of granting
agencies.
• Locating possible grantors is a time consuming task (except in cases
where it is announced) but in the long run it will yield the greatest
benefits.
• Even if you have the most appealing research proposal in the world, if
you don’t send it to the right institutions, then you’re unlikely to receive
funding.
• There are many sources of information about granting agencies and
grant programs.
• Library, Google Search, with key words help you to find funding
agencies, inputs from colleagues/experts …..
 What is Project (Wikipedia)
• Research is "creative and systematic work undertaken to
increase the stock of knowledge".

• It involves the collection, organization and analysis of information to

increase understanding of a topic or issue.

• A research project may be an expansion on past work in the field. To test

the validity of instruments, procedures, or experiments, research may
replicate elements of prior projects or the project as a whole, but bring ? .

• Project Proposal : May be basic science, applied science, transdisciplinary

translational Research, technology dev., new discoveries, inventions ............

- Has an established objective and realization within costs & quality

- Has a defined time frame – Start and End date (short or long term)
- Involves doing something never been done before
- Specific and finite activity that gives a measurable and observable result (input & output)
- May be collaborative/individual initiative that is carefully planned to achieve a particular aim

• Skills for Writing Proposals : Technical & Management Skills

Project Proposal: Important Steps
Project Title : Spend time developing the best possible title
• The title summarizes the main idea or ideas of your study.
• A good title contains the fewest possible words that adequately
describe the contents or purpose of research.
• Indicate accurately the subject and avoid using abbreviations.
• Use words that create a positive impression and stimulate reader
interest.
• May reveal how the paper will be organized.
• Is limited to 10 to 15 substantive words.

• Do not include "study of," "analysis of" or similar constructions. Titles

are usually in the form of a phrase, but can also be in the form of a
question
Project Proposal: Important Steps
Abstract/Executive Summary
• Aim to catch your recipient's attention and encourage them to read on.
• Executive summary should include details about:

▪ Purpose for developing the proposal

▪ The problem your project solves
▪ How your project solves the problem
▪ Your project's intended impact - Beneficiaries – who and how they will
be benefitted
• The length of your executive summary will depend on your project's
complexity.
• Make your executive summary concise to keep it engaging and convincing.
• Create linear flow of logic when linked, one after another.
Introduction /Background to the Project
• What is the topic? Why is this topic important?
• What is the problem your project addresses?
• What is already known about this problem?
• Who has addressed this problem before/what research is there?
• Why is past research insufficient at addressing this problem? Will you be
using new approach? What practical or valuable problems will it help solve?
• The introduction should cover the key elements of your proposal, including
a statement of the problem (the critical issue that your research seeks to
address), need, the purpose of research and significance/relevance
• Literature reviews should be selective and critical, not exhaustive.
Reviewers want to see your evaluation of pertinent works (give credit where
due).
• List the publication details from the last 5 years which should be relevant to
the proposed work. (Never include predatory journals)
 Objectives of the Research Proposal
• Research objectives is crucial for conducting a successful research
• Make sure to clearly describe what you aim to achieve through your
research - Pinpoint the major focus of your research
• Break down your research focus into research objectives
• Writing strong research objectives is to write your objectives as
concisely as you can and be realistic.
• Write only a few specific research objectives (5 or 6), Use action
verbs – Assess, Determine, Calculate, Compare, Explain, Describe
• Writing strong research objectives is to use the SMART format
(Specific, Measurable, Achievable, Relevant and Time-based)
• Research objectives should identify the anticipated outcomes of the
research and should match up to the needs identified in the
statement of problem.
 Research Methodology
• This part should be written properly and organized logically since
you are not yet conducting the actual research. However, it must
build confidence among your readers that it is something worth
pursuing.
• It gives rightfulness to your research. An unreliable or erroneous
methodology produces unreliable or erroneous results.
• The reader of your research expects you to have followed accepted
practices so that the conclusions you reach are valid.
• Needs to be repeatable, meaning anyone who uses the methods you
write about should reach the same conclusions you reached.
• If any part of your process is outside of the realm of usual practices
in your field, clarify your choice
• The underlying purpose here is to convince the reader that your
research design and suggested analytical strategies will properly
address the problem/s of the study.
 Research Methodology
• Research design and methods should be directly connected to
the particular objectives of your research

• How are you going to achieve those outcomes? What methods

will you use?

• How will you measure or recognize your project’s

achievements?

• How can you be sure that your project will productively

respond to the need or problem you have identified?
 Work Program/Plan
• Schedule the work elements listed in the methodology in a sequence
indicating the role of each associate.
• Standard project management techniques like flow chart, gnat chart or
PERT network
• Specify facilities available and additional resources needed and the
method of acquiring resources.
• Duration of the Project: Specify time frame activity wise. Give
milestones for each objective (2 to 3 years preferred).
• Elaborate how the work will be managed. if the project is multi-
location or multi-institutional in nature.
• Most funding agencies require list of equipment with detailed
specifications, time, and schedule for procurement and finally the
actual users of the equipment.
• The facilities available at the host institution to be mentioned clearly.
• Give details of training needed linking their relevance to the project.
 BUDGET
Most funding agencies need cost details in the following heads
✓ Establishment (Personnel - salaries)
✓ Recurring and Non-recurring
(Equipment’s, Travel, Field Visits, Data Analysis ……………)
✓ Miscellaneous
✓ Overhead
• to be specified year-wise.
• Justify costly equipment and other facilities.
• Try to include reasonable inflation while calculating year-wise budget.
• Give summary budget.
• Specify budget for each partner if they come from different places and
administrative units
• Budgets should match the details given in the work program.
• Avoid over budgeting, Provide Justification
 Conclusion
• One of the best ways to conclude your research
proposal is by presenting a few of your anticipated
outcomes.
• This will allow your evaluator to further assess the
significance and validity of your project.
• It will also indicate that you have thought ahead and
considered the potential outcomes and implications
of your research.
 Supporting Documents
• You may need to provide a range of supporting materials at the
end of your proposal—usually in the form of appendices.
• These might consist of additional records, endorsements,
personnel bios of investigators/collaborators, letters of support
from allied organizations
• More Publications in international journals indexed by
SCOPUS, Web of Science and more citations will help the
researchers or faculty members to fetch the grant
• All of this documentation should be clearly related to your
proposal and may be requested by the granting institution.
 Some General Tips
• Begin early and Apply early
• Don’t forget to include a cover letter with your application
• Answer all questions (Pre-empt all unstated questions)
• If rejected, revise your proposal and apply again
• Give them what they want. Follow the application guidelines exactly.
• Be explicit and specific.
• Be realistic in designing the project.
• Make explicit the connections between your research questions and
objectives, your objectives and methods, your methods and results,
and your results and dissemination plan.
• Follow the application guidelines exactly (it is very, very important.)
• Revisit the complete proposal twice before submitting.
• Stay focussed and avoid jargons
 Some General Tips
• Your ability to find funding for your research program will be a major
measure of your success in the University.
• Look for forging collaborative research with reputed/premier research
institutions, R&D organizations and industries – both at state/national
and international level.
• Collaboration may be considered with individuals, who have track record
of good research and also who can help in preparation of research
proposals.
• Young researchers shall be teamed-up with experienced and seasoned
researchers both within and outside researchers, even in the publication of
research papers in reputed and peer reviewed journals.
• Well in advance write general proposal, which is sometimes called a
“white paper.” Before you submit proposals to different grant programs,
you will tailor a specific proposal to their guidelines and priorities.
• Suggest "benevolent" reviewers (i.e. people who you know or who know
you)....
Research Proposal : A Typical Flow Chart
Identify Needs Status, New Ideas
What? Why?
and Focus How?
Literature on Idea
Review the Discussion with
Literature on Topic
Literature Liter. on Approach Peers & Experts
Develop
Why ? & How ? Clear Approach
Objectives
Research Design, Collaboration:
Design the Study and
Develop Methodology
Research Procedures Intra & Inter
Data: Type & Availab. Ethics, IPR
Project Planning Assess University Relationship between
Time Frame & Est. Budget Infrastructure parameters
Peer Review New Equipment . User Needs-Local, State …..
Develop General …National & International
Fine Tune the Proposal
Look for Funding
Proposal & Undertake Based on Funding Agency
Solicited, Unsolicited & Guidelines
Preliminary Study
Review and Application
Submission before
Deadline
 DST : Scientific Programmes
• Scientific & Engineering Research
• Technology Development
• International S&T Cooperation
• S&T for Socio Economic Programme
• Technology Missions Division
• Women Scientists Programs
 Department of Science and
Technology (DST), GOI
Scientific & Engineering Research
• Mega Facilities for Basic Research
• Innovation in Science Pursuit for Inspired Research (INSPIRE)
programme
• R&D Infrastructure (FIST, SAIFs, PURSE, SATHI, STUTI)
• Science and Technology of Yoga and Meditation (SATYAM)
• Programme for Science Students
• Swarnajayanti Fellowships
• National Science & Technology Management Information System
(NSTMIS)
• Science and Engineering Research Board (SERB)
• Cognitive Science Research Initiative (CSRI)
• VAJRA (Visiting Advanced Joint Research) Faculty Scheme
 DST: R&D Infrastructure (FIST, SAIFs,
PURSE, SATHI, STUTI)
• Aims to strengthen the S&T infrastructure of the country by fostering well-
equipped R&D labs in the academic/ research institutes/ universities
• Strong culture of research collaboration between institutions and across
disciplines.
• Five schemes aiming at establishment of R&D labs, centres, upgradation of
research facilities orienting towards creating a self- reliant India.

▪ FIST (Fund for Improvement of S&T Infrastructure in Universities and

Higher Educational Institutions)
▪ PURSE (Promotion of University Research and Scientific Excellence)
▪ SAIF (Sophisticated Analytical Instrument Facilities)
▪ SATHI (Sophisticated Analytical & Technical Help Institutes)
▪ STUTI (Synergistic Training program Utilizing the Scientific and
Technological Infrastructure)
 Science and Engineering Research
Board (SERB), DST, GOI
• SERB supports research in frontier areas of Science and Engineering.
A regular faculty / researcher in an academic / research institution
can seek research support to carry out the research.
• Provides special attention to young scientists below the age of 35
years (relaxable by 5 years in the case of SC/ST/OBC, woman and
physically handicapped) to undertake independent research in
newly emerging and frontier areas of science and engineering.
• Funds awarded based on novelty, investigator competence to
execute the project and the infrastructure available (even to Private
University)
• The Board funds only one project to an Investigator at a time, except
in cases where calls for proposals are invited for specific
areas/schemes/programs.
 SERB: Schemes & Programs
• Core Research Grant (CRG)
• National Mission on Interdisciplinary Cyber-Physical Systems
(NM-ICPS)
• SERB Science and Technology Award for Research (SERB-STAR)
• Industry Relevant R & D (IRRD)
• Empowerment and Equity Opportunities for Excellence in science
(EMEQ)
• Intensification of Research in High Priority Area (IRHPA)
• Start-up Research Grant (SRG)
• MATRICS
• International Travel Support (ITS)
• Assistance to Professional Bodies & Seminars/Symposia
• Start-up grant for Young Scientists(YSS)
• Accelerate Vigyan
 All India Council for Technical
Education (AICTE)
• Faculty Development Schemes
• Student Development Schemes
• Institutional Development Schemes
• Research & Innovations Development Schemes
• General Schemes
 AICTE: AICTE-RESEARCH
PROMOTION SCHEME (RPS)
• Promotes Research in identified thrust areas - in Technical Education.
• Aimed to create research ambience by promoting research in engg.
sciences and innovations in established and newer technologies
• The limit of funding is Rs 25 Lakh for project duration of 3 years.
• The proposal should include a specific project theme with a clear
statement of the objectives, details of equipment and other research
facilities proposed to be acquired and the expected deliverables from
the project.
• One proposal from one PI (Principal Investigator) will be sanctioned
at any point of time within the duration of project.
• Maximum 2 number of proposal acceptable to be approved per
institute
Faculty Development Programme
(FDP)
• Provide financial assistance to facilitate up-gradation of knowledge, skill and
opportunities for induction training to teachers
• Award Faculty Development Programme (FDP) to cover areas such as
technical education policy, new concepts, methods and techniques, theory
and skills development and up gradation of pedagogy educational
technology, motivation, communication skills, management and other
relevant issues to keep pace with the changing scenario in Technical
Education.
• To enhance the teaching and other skills of the faculty. And to make them
aware about modern teaching tools and methodologies.
• It provides an opportunity to acquire knowledge about current technological
developments in relevant fields. It will not only promote the professional
practices relevant to technical education but also motivates the faculty to
achieve competitive teaching and learning environment, thus channelizing
development with respect to academic qualifications and personal matters.
 AICTE - Professional
Development Scheme
• Providing financial assistance to regular faculty of the AICTE to
present research papers – both international & India
• Aim : to update themselves to the global changes in the concerned
field of specializations through paper presentation in conference
abroad & within India.
• Eligibility : Full time regular faculty from the AICTE approved
University Department (with at least 5 years of existence).
• The applicant must not have availed the grant under this scheme
during the last three years.
• Applicant must have received acceptance of paper from the
organizer of conference. Limit of Funding Up to Rs.1.5 Lakhs.
Council of Scientific & Industrial
Research (CSIR)
• CSIR known for its cutting edge R&D knowledge base in diverse S&T
areas
• CSIR has a dynamic network of 37 national laboratories, 39 outreach
centers, 3 Innovation Complexes, and five units with a pan-India presence.
• CSIR covers a wide spectrum of science and technology – from
oceanography, geophysics, chemicals, drugs, genomics, biotechnology
and nanotechnology to aeronautics, instrumentation, environment, IT …
• It provides significant technological intervention in many areas
concerning societal efforts, which include environment, health, drinking
water, food, housing, energy, farm and non-farm sectors.
• CSIR filed about 225 Indian patents and 250 foreign patents per year
during 2015-20. CSIR is a leader in filing and securing patents worldwide.
• CSIR is ranked 37th among 1587 government institutions worldwide and
is the only Indian organization among the top 100 global government insti.
Council of Scientific & Industrial
Research (CSIR)
• Research Schemes
• Emeritus Scientist
• Sponsored Research Schemes
• Grant for Journals

• The Directors of CSIR laboratories may invite applications for

research grants in specific areas of interest to their respective
laboratories. They will forward these to the CSIR HRD Group.

• Schemes in which collaboration has been indicated between

industries and CSIR laboratories or between universities and CSIR
laboratories or among industries, universities and CSIR
laboratories, will receive support on priority basis on the approval
of DGSIR.
 List of Funding Agencies:
National and International
Agencies
• AICTE - All India Council for Technical Education -www.aicte-india.org
• CSIR - Council of Scientific and Industrial Research -www.csirhrdg.res.in
• DAE-BRNS - Department of Atomic Energy - www.dae.gov.in
• DAE-NBHM - National Board for Higher Mathematics -
www.nbhm.dae.gov.in
• DBT - Department of Biotechnology - www.dbtindia.nic.in
• DOE - Department of Electronics - www.dde.nic.in
• DRDO - Defence Research and Development Organization -
www.drdo.gov.inz
• DRDO-AR&DB - Aeronautics Research and Development Board -
www.drdo.gov.in
• DRDO-DMRL Defence Metallurgical Research Laboratory -
www.drdo.gov.in
• DRDO-LSRB - Life Sciences Research Board - www.drdo.gov.in
 List of Funding Agencies: Cntd,..
• DSIR - Department for Scientific and Industrial Research - www.dsir.gov.in
• DST - Department of Science and Technology - www.dst.gov.in
• ICMR - Indian Council of Medical Research - www.icmr.nic.in
• IGCAR - Indira Gandhi Centre for Atomic Research - www.igcar.ernet.in
• INSA - Indian National Science Academy - www.insaindia.org
• IREDA - Indian Renewable Energy Development Agency Limited -
www.ireda.gov.in
• ISRO - Indian Space Research Organization - www.isro.org
• ISTE - Indian Society For Technical Education - www.isteonline.in
• IE -The Institution of Engineers (India) - www.ieindia.org
• IET - The Institution of Engineering and Technology - www.theiet.org
• IMD - India Meteorological Department - www.imd.gov.in
• MCIT - Ministry of Electronics and Information Technology -
www.meity.gov.in
 List of Funding Agencies: Cntd,..
• ME (MHRD) – Ministry of Education -www.education.gov.in (Ministry of
Human Research Development/ - www.mhrd.gov.in)
• MNRE - Ministry of New and Renewable Energy - www.mnre.gov.in
• MoEF - Ministry of Environment and Forests - www.moef.nic.in
• MoES - Ministry of Earth Science - www.moes.gov.in
• MOWR - Ministry of Water Resources - www.wrmin.nic.in
• MOCIT - Ministry of Communications & Information Technology -
www.deity.gov.in
• MFPI - Ministry of Food Processing Industries - www.mofpi.nic.in
• NCERT National Council for Economic Research and Training -
www.iccw.vsnl.net.in
• NJB - National Jute Board - www.jute.org
• PCRA - Petroleum Conservation Research Association - www.pcra.org
• RSFC - Research Scheme on Flood Control - www.rsfc.vsnl.net.in
• RSOP - Research Scheme on Power - www.cbip.org
 List of Funding Agencies: Cntd,..
• SDC - SAARC Documentation Centre - www.sdc.gov.in
• SERC - Science and Engineering Research Council - www.serc-dst.org
• TAAS - Technology Absorption and Adaptation Scheme - www.taas-dst.org
• UGC - University Grants Commission - www.ugc.ac.in
• VSSC - Vikram Sarabhai Space Centre - www.vssc.gov.in
• W2E -Waste to Energy - www.w2es.com
• British Petroleum (BP)/Bird life International/ Conservation
International/Fauna and Flora international/ Wildlife conservation
society
• CIFAR- Canadian Institute for Advanced Research
• DST-SERB -The Science and Engineering Research Board (SERB) has its
origins from the erstwhile Science & Engineering Research Council (SERC),
DST, Government of India
• IBRO - International Brain Research Organisation
• DBT- Wellcome Trust India Alliance
 List of Funding Agencies: Cntd,..
• Alexander Humboldt Foundation
• American Cancer Society
• DFG – German Research Foundation
• EARTHWATCH Institute US/International
• James McDonnell Foundation (USA)
• SFARI - Simons Foundation Autism Research Initiative
• SreePVF - Sree Padmavathi Venkateswar Foundation
• HHMI/ Gates/ Wellcome Trust/ Calouste Gulbenkian Foundation
• HFSP - Human Frontier in Science Program
• Indo-US Science and Technology Forum
• IFCPAR- Indo French Centre for the Promotion of Advanced Research
• INNO INDIGO Partnership Programme
• Indo-US Science and Technology Forum
• National Geographic
 List of Funding Agencies: Cntd,..
• NSF-SERB
• Lady Tata Memorial Trust
• Ratan Tata Trust and Navajbai Ratan Tata Trust
• INSA-CSIR-DAE/BRNS-DOS/ISRO
• ICMR - Indian Council of Medical Research
• TWAS - The World Academy of Sciences
• Indo-US Science and Technology Forum
• EMBO - European Molecular Biology Organization
• EMBO/Wellcome Trust/DBT India Alliance
• DST/Royal Society and Department for Business, Innovation & Skills (BIS)
UK
• IFCPAR - Indo-French Centre for the Promotion of Advance Research
• Company of Biologists
• Matsumae International Foundation (Japan)
 List of Funding Agencies: Cntd,..
• IUSSTF - Indo-U.S. Science and Technology Forum
• INSA - Indian National Science Academy
• DST/BELSPO, Belgium
• DBT/ MCTI (Brazil)
• DAAD, Germany
• DST/ DAAD (Germany)
• CEFIPRA-DST
• Cambridge –India
• BBSRC (UK) - Biotechnology and Biological Sciences Research Council
• Wellcome-DBT India Alliance
• DST/RCN, Norway
• DBT/DFG, Germany
• DBT/EU, Europe
• DBT/DIIS, Australia
 List of Funding Agencies: Cntd,..
• NSF-SERB
• BES - British Ecological Society
• Rolex Awards
• CEFIPRA-BIRAC
• BIRAC - Biotechnology Industry Research Assistance Council
• Wellcome Trust (UK)
Happy to see you all
National Education Policy
--A Bird View--

Murali S

[email protected]
When so many policies are in India
there like defence, labour, income
tax, land, revenue, excise, law and
order, agriculture etc etc etc

Why not for education

 NEP is good
Do not expect too much
reason :
1. Mind set
2. Bureaucracy
3. Infrastructure
4. Political
• The world is undergoing rapid changes in the knowledge landscape.

• With the rise of big data, machine learning, and artificial intelligence, many unskilled
jobs worldwide may be taken over by machines, while the need for skilled labour,
particularly involving mathematics, computer science and data science, in conjunction
with multi-disciplinary abilities across the sciences, social sciences and humanities,
will be in rapidly increasing demand.

• With climate change and rapid depletion of natural resources, there will be a sizable
shift in how we meet the world’s energy, water, and sanitation needs, again resulting
in the need for new skilled labour, particularly in biology, chemistry, physics, and
climate science.

• There will be a growing demand for humanities and art, as India moves towards
becoming a developed country and among the three largest economies in the world
World-class institutions of ancient India, such as Takshashila, Nalanda,
Vikramshila and Vallabhi set the highest standards of multidisciplinary
teaching and research and hosted scholars and students from across
backgrounds and countries.

The Indian education system produced scholars like Charaka and Susruta,
Aryabhata, Bhaskaracharya, Chanakya, Madhava, Patanjali, Panini and
Thiruvalluvar, among numerous others.

They made seminal contributions to world knowledge in diverse fields, such

as mathematics, astronomy, metallurgy, medical science and surgery, civil
engineering and architecture, shipbuilding and navigation, yoga, fine arts,
chess, and more. Indian culture and philosophy has had a strong influence on
the world
The teacher and the teacher’s
condition must and will be at the
centre of these changes.

The new education policy must help

reinstate teachers, at all levels, as the
most respected and essential
members of our society
 Earlier
• National Policy on Education 1986, Modified in 1992 (NPE 1986/92).

• A major development since the last Policy of 1986/92 has been the
Right of Children to Free and Compulsory Education Act 2009
I SCHOOL EDUCATION

II HIGHER EDUCATION

III OTHER KEY AREAS OF FOCUS

IV MAKING IT HAPPEN
ECCE- Early Childhood Care and Education SDG- Sustainable Development Goal
The principles of this Policy
flexibility, no emphasis on
hard conceptual on creativity
multi-
separations understanding and critical
disciplinary
between arts (rather than rote thinking
and sciences learning and
learning-for-exams)

ethics and
regular respect for
human &
life skills formative diversity
Constitutiona
l values assessment for
learning
respect
for the teachers and
local equity and
inclusion resource faculty as the
efficiency heart of the
learning process
light but tight
oversight and
regulatory ‘autonomy,
system to ensure good
integrity and outstanding continuous
governance and
transparency research policy-making
empowerment
 Finally
education is a public service and not a
commercial activity or a source of profit; access
to quality education must be considered a
fundamental right of every citizen;

substantial investment in a strong, vibrant

public education system - as well as the
encouragement and facilitation of true
philanthropic private participation.
Schools
• Click here
 Few Shot Learning: A
Perspective on Face Recognition

V.N. Manjunath Aradhya

Dept. of Computer Applications
JSS Science and Technology University, Mysuru
[email protected]
98868 96108
 The Talk
• Overview on Concept Learning
• Few Shot Learning
• Motivation and Challenges
• Learning Problems and Core Issue
• Case Study
• Future Scope
How Humans do?
Human Cognition?
 Human Cognition?
• Human Cognition is permeated by self-control: the ability to engage in
complex, goal-oriented behaviors rather than just react to the moment at
hand.
– Includes: Thinking, knowing, remembering, judging, and problem-
solving.

Information,
Thoughts
Sensations

Emotions,
Memories
Why we have to learn (Humans)?
 Why Learn (Machine)?
• Used When:
– Human Expertise Doesn’t Exist (Moon /
Mars)
– Human are unable to explain their expertise
(Speech)
– Solution needs to be adapted to particular
cases (Biometrics)
 Learning (Machine)
• Three general categories of learning:

– Supervised Learning
– Unsupervised Learning
– Reinforcement Learning
Real Examples
 Supervised

• Cortana / Speech Automated System

• Various Features (past records) – predicts the
winning %
• OCR system – Convert Handwriting to Text
• Spams
• Biometric Attendance
 Unsupervised

• Party - Classify them “on the go”

• Discoveries – NASA
• Cricket - Classify based on different criterion
• Survey of 500 Questionnaire – trim down the
questionnaire (PCA)
 Concept Learning
The term concept learning is originated in psychology, where it refers to the human ability to
learn categories for object and to recognize new instances of those categories.

• By age of 6, we could recognize more than 100’s of categories

of objects and keep learning.
• Humans have ability to learn concepts from few examples.
• This has resulted from the unique properties exhibited by
humans while learning. Some of these properties are:

– 1) Compositionality – Observing an object, humans can

identify it parts & subparts.
– 2) Causality – Causality explains how a new concept can be
learned hierarchically from related concepts and this is one
of the reasons for fast learning by humans.
– 3) Learning to learn – This refers to the fact that inductive
priors while learning a particular task can be shared while
learning another task.
 One Shot Learning?

• Ability to learn from a single instance is

something unique to the human species and
One-shot learning algorithms try to mimic
this special capability.
• One-shot learning aims to train image
classifiers over the dataset with only one
example / few examples per category
Few (Low) Shot Learning (FSL/LSL)
• Learning with limited data.
• It is different from standard supervised
learning.
• Goal: Learn to Learn
– Not to build model, which recognize the images in
the training set and then generalize to the test set.
– similarity and difference between objects
 Applications Law Enforcement, Forensics
(Limited Data)
Autonomous Driving

Language Acquisition and Understanding

Diagnosing Developmental Disorders in

Children
Medical Field

Manufacturing
 Motivation
• Test base for learning like human
• Scarce cases
• Reduce Data Collection effort and
Computation
Case 1
Case 2
Can machine predict the correct one?
Case 3
 What we do in
traditional ML?
Finally, What’s FSL?
0.9 Not in
0.3 the DB

0.05

0.05

0.1
0.2
 Traditional Supervised Learning Vs Few Shot Learning

Test Samples not seen before Test

Test Samples are from known class

Training Set

Support Set

Test Test Samples not seen before

Test Samples are from unknown class
 k-way n-shot support set

k-way: the support set has k classes 5-way

Support Set
n-shot: every class has n samples

5-way 1-shot 3-shot

 Contd…
• Few-shot learning is typically measured by its
performance on n-shot, k-way tasks.
• These are run as follows:
– A model is given a test sample belonging to a new, previously unseen class
– It is also given a support set, S, consisting of n examples each from k different
unseen classes
– The algorithm then has to determine which of the support set classes the
query sample belongs to
 Predication Accuracy
Accuracy Accuracy

# of shots
# of ways
 Few-Shot Variations
In general, researchers identify four types:

• N-Shot Learning (NSL)

N-Shot Learning is seen as a more broad concept than all the others. It means
that Few-Shot, One-Shot, and Zero-Shot Learning are sub-fields of NSL.

• Few-Shot Learning
• One-Shot Learning (OSL)
• Less than one or Zero-Shot Learning (ZSL)
 FSL Approaches
Define:
• A training (support) set that consists of:
– k class labels
– n labeled images for each class (a small amount, less
than ten samples per class)
• Q Test images
We want to classify Q test images among the k classes.
The n * k samples in the training set are the only examples
that we have.
Support Set Test Images

Important step in an FSL task is to gain experience from

other, similar problems.
Also, characterized as a Meta-Learning problem.
 • Meta-learning:
– also known as “learning to learn”, intends to design models that
can learn new skills or adapt to new environments rapidly with a
few training examples
– Applications: guiding policies by natural language, automation of
factory operation, learning optimizers etc.
– Ex: A robot completes the desired task on an uphill surface test
even though it was only trained in a flat surface.

*Meta-learning methods can be used to deal with the OSL/FSL problem. Meta-
learner is taken as prior knowledge to guide each specific OSL/FSL task.
 Approaches
• Prior knowledge about similarity:
– FSL techniques enable ML models to separate two classes that are not present in the
training data(Siamese, Triplet)

• Prior knowledge about learning:

– ML models use prior knowledge to constrain the learning algorithm to choose
parameters that generalize well from few examples. (LSTM, Reinforcement etc)

• Prior knowledge of data:

– ML models exploit prior knowledge about the structure and variability of the data,
which enables construction of viable models from few examples. (Pen stroke
model, Analogies etc.)
 Basic Idea & Example
• First, learn similarity function from large dataset.
• Then, apply similarity function for prediction
– Compare the test with the sample in the support set
– Find the sample with highest score
Siemese Network
 Applications
• Computer Vision
• NLP
• Robotics
• Acoustic Signal Processing
• Other Applications
 Core Issue
• In ML – there are prediction errors and one cannot obtain perfect predictions.
• Core issue of OSL/FSL is based on error decomposition
– Empirical Risk Minimization – Total error can be reduced by attempting
perspectives of data, model, and algorithm
– Unreliable Empirical Risk Minimizer - when there is sufficient training data
with supervised information , the empirical risk minimizer can provide a good
approximation
– Indeed, this is the core issue of OSL/FSL supervised learning, i.e., the
empirical risk minimizer is no longer reliable.
 Taxonomy
• To overcome the problem of having an unreliable empirical risk minimizer
in OSL/FSL, prior knowledge must be used.
• Existing FSL works can be categorized into the following perspectives:
• Data
– Use prior knowledge to augment Data and increase the number of samples from I to
Il, where Il ≫ I. Accurate empirical risk minimizer can be obtained
• Model
– In order to approximate the ground-truth hypothesis h^, the model has to determine
a hypothesis space H containing a family of hypotheses h’s, such that the distance
between the optimal h∗ ∈ H and h^ is small.
– OSL/FSL methods manage to learn by constraining H to a smaller hypothesis space
H1 via prior knowledge in Experience. The empirical risk minimizer is then more
reliable, and the risk of overfitting is reduced.
• Algorithm
– Alter search strategy in hypothesis space by prior knowledge.
 References
• One-shot learning of object categories, IEEE Trans. on PAMI, 2006
• One-Shot learning for Custom Identification Tasks: A Review, Procedia
Manufacturing, 2019.
• Generalizing from a Few Examples: A Survey on Few-Shot Learning, arXiv,
2020.
• One Shot Cluster based Approach for the Detection of COVID-19 from Chest X-
Ray Images, Cognitive Computation, 2020 (In Press).
• One-shot Learning for iEEG Seizure Detection Using End-to-end Binary
Operations: Local Binary Patterns with Hyperdimensional Computing, IEEE
BioCAS 2018.
• Face Recognition - A One-Shot Learning Perspective, 15th International
Conference on Signal-Image Technology & Internet-Based Systems (SITIS), 2019.
• One-Shot Digit Classification Based on Human Concept Learning, ICCIS, 2019
Life skills and
Personality
Development

Geetha Kiran A.
BEING PROFESSIONAL OR NOT?
Let’s take a little scientific test
 Correct answer: Open the door and put him in.
 How do you put a giraffe in a refrigerator?
 Correct answer: Take out the giraffe and put in
 How you put an elephant in a refrigerator?
the elephant.
 The Lion King is hosting a conference. All of the
 Correct answer: The elephant, he is still in the
local animals are in attendance but one. Who is refrigerator.
not present?
 Correct answer: You swim; all of the crocodiles
 You are slated to be the final speaker at the
are all ready at the conference
conference. Unfortunately, to get there you have
to swim a river inhabited by crocodiles. What do
you do?

A professional looks for the simplest solution

(#1). Keeps things simple.
(#2). Recalls facts (the elephant is still in the refrigerator and thus can’t go to the conference.
(#3). And, uses his/her deductive skills in problem solving (the crocodiles are at the conference).
 VARIABLES INFLUENCING
INDIVIDUAL BEHAVIOR

The Person
• Skills & abilities The Environment
• Personality • Organization
• Perceptions • Work group
• Attitudes • Job
•Values • Personal life
• Ethics

Behavior
Definition of
Personality

Personality - A relatively stable set

of characteristics that influences
an individual’s behavior
PERSONALITY CHARACTERISTICS
IN ORGANIZATIONS

Locus of Control
Internal External
I control what People and
happens to me! circumstances
WHO IS MOST LIKELY TO . . .

Low-self High-self
monitors monitors
Get promoted/Not
Accomplish tasks, meet other’s
expectations, seek out central positions ✓
in social networks
Exposure
Self-promote ✓
Make a geographic
move
Demonstrate higher levels of managerial ✓
self-awareness; base behavior on other’s
cues and the situation
PERSONALITY
TRAITS
WARM –UP: SELF AWARENESS ACTIVITY
CAN A LIFE SKILL BE TAUGHT?
Yes & No.
I say a ‘no’ because finally it is up to the individual concerned;
& a ‘yes’ because the environment greatly influences a person’s mind, behaviour and actions.
EMPATHY

Mahabharata - Yudishtira is the most important

character because he teaches us the basic and fundamental
value of compassion to be a human being.

Without empathy, Empathy can help

Understand and Empathy is the our communication us to accept others,
care about other ability to imagine
with others will who may be very
peoples’ needs, what life is like for amount to one-way different from
desires and feelings. another person. traffic. ourselves.
 Well lets play!!!!!

If it’s true
Everyone The person of anyone
stand in a That is true
in the middle else, they
circle with to say about quickly have
one person themselves
EMPATHY something to switch
in the middle places
THE “I AM” CIRCLE
 CRITICAL THINKING

 Critical thinking can be formally understood as

`the intellectual process that helps a human progress toward the objective truth’.
For us to fulfill that higher and more impactful potential within us, we need to move from the position of
`prove I am right’ that comes from our ego to `seek the truth’ which is beyond personal limits.
This journey unlocks excellence in decision making - leading, in turn, to personal fulfillment.
 Creative 1
2
3
Thinking Elaboration (building Fluency (generating Flexibility (shifting
on other ideas). new ideas) perspective easily),

Creative Thinking is a novel way

of seeing or doing things that are
characteristic of
4

Originality
(conceiving of
something new)
 ‘SAY WHAT YOU SEE’ GAME
To set up, give each person a piece of paper and a pencil:
 Draw a picture of the surroundings.

 Take a deep breath.

 Look to the right and say what you see.

 Look to the left and say what you see.

 Look in front and say what you see.

 Look behind and say what you see.

 Look up and say what you see.

 Look down and say what you see.

Flip the paper and draw a picture of everything you can remember about where they are.

Being more mindful about surroundings

• Decision-making helps us to deal
constructively with decisions about our
lives.
• This can have consequences for health.
Decision
• It can teach people how to actively make Making
decisions about their actions in relation to
the health assessment of different options
and, what effects these different decisions
are likely to have.
HUMAN KNOT

Stand in a circle, and hold hands with the person

who isn’t directly next to you.
Now untangle the knot and form a perfect circle
— without letting go of anyone's hand.
 PROBLEM SOLVING

 Problem-solving helps us to deal constructively with problems in our lives.

 Significant problems that are left unresolved can cause mental stress and give rise to accompanying physical strain.

 Problem-solving is the process of identifying and resolving conflicts or problems.

 It involves breaking a problem down into its component parts, thinking about possible solutions, and then choosing the
best one.
 Problem-solving is an important skill for everyone because it helps us identify and solve problems in our lives, work,
and communities.
INTERPERSONAL RELATIONSHIP

I E
• Gentle • Interested • Value • Easy

G V
 STORY TIME GAME

 Lets start a story – with a sentence

 It now goes in a circle, with each of you repeating the sentence that has just been said and then adding a new sentence
onto the story.
 This exercise is good at exposing the differences in the Sensing and Intuitive thinking style.

 Sensors typically will continue the story in a linear fashion whereas Intuitives will play around with ideas, adding their
own interpretations as to how each of the characters might think, act and feel.
EFFECTIVE  Draw something by following my verbal directions.
COMMUNICATION  Cannot ask questions.
Instructions :
 Draw a small circle near the top of the page.
 Now draw a vertical line from the circle to the middle of the page.
 Now draw two lines from the circle angled down towards the middle of the
page.
 Now draw two lines separating from the end of the single vertical line each of
which angles down towards the bottom of the page.
 At the end of the angled lines stemming from the circle, draw five small lines.
 At the end of the angled lines stemming from the vertical line, draw an oval at
the base of each line.
 Look at the picture I have constructed and compare.
COPING WITH STRESS

 Is a fact of life for most people.

 You may not be able to get rid of it.

 But you can find ways to lower it.

STRESSORS
 QUICKEST SOLUTION

 Involve your senses.

 Everyone is unique so you need to find what works

for you.

 You may need a combination of techniques to be

most effective.

 Keep trying.
 STRESS BUSTER
 try this one from The Canadian Mental Health Association.

 https://cmha.ca/whats-your-stress-index
 COPING WITH EMOTIONS

Match the pictures on the left with the emotions listed on the right.
Pictures may have more than one answer and you may add emotions that are not
listed here.

Our feelings are a wonderful

barometer of our well-being.
TIME MANAGEMENT
 Say “no” to more unnecessary tasks

 Prioritize (ensure your health is a priority)

 Map it out so you can coordinate your time-write

down everything (if you are unable to pencil in
new tasks- do not agree to take them on)
 Stay organized

 Know your limitations

 Review your schedule weekly and adjust and

prioritize accordingly
 Plan ahead so you are not scrambling to
complete tasks
 TIME MANAGEMENT

 Game -1 – How Long is a Minute?

 A minute is 60 seconds.

 This a good game to start with and it goes like this:

 Put your phones away and remove your watches.

 Close your eyes.

 Stand up and then open your eyes when a minute has passed.
CORE LIFE SKILL STRATEGIES AND TECHNIQUES
THE KEY WORD IS….

Balance
 Human Gait
Recognition: Review
and Challenges

Dr. H.S.Nagendraswamy
Professor

DOS in Computer Science

University of Mysore
Manasagangothri
Mysore-6
Every interaction is an opportunity to learn

Only if we are interested in Improving

rather than Proving…..

Outline
 Introduction to Biometrics
 Gait Biometric
 Gait Recognition System
 History of Gait Recognition
 Gait Databases
 Gait Recognition Approaches
 An Example Symbolic Representation Technique to GR
 Summary and Conclusions
 References
Introduction
 Biometric refers to
 Identifying or verifying humans.
 Physiological characteristics
 Fingerprints, face, iris, hand etc.
 Behavioural characteristics
 Signature voice/speech, gait, etc.
 Biometric systems operate as either
 Identification' system or a verification system.
 Identification
 One to many comparisons.
 Verification
 One to one comparison.
 All biometric system performs
 Recognition : again know a person who has been previously enrolled.
To be considered a biometric qualification criterion, a candidate feature must meet the
following conditions

 Universality: each person should have the characteristic.

 Distinctiveness: any two persons should be sufficiently different in terms of the
characteristic.
 Permanence: the characteristic should be sufficiently invariant over a period of
time.
 Collectability: the characteristic can be measured quantitatively.

Issues to be considered in practical biometric system

 Performance: Refers to the achievable recognition accuracy and speed.

 Acceptability: Indicates the extent to which people are willing to accept the use of
a particular biometric identifier in their daily lives.
 Circumvention: Denotes the easiness the system is fooled through fraudulent
methods.
Gait Biometric

 Gait: A person’s manner of walking.

 Gait Recognition : Identifying individuals by the way they walk.
 Gait biometric features
 Different from traditional biometrics.
 Incorporate both spatial as well as temporal components.
 It is unique in every person.
 It can be captured at a distance.
 It does not need individual cooperation.
 It can be used even with low resolution.
 It is difficult to hide, steal or fake.
 It can be used for some early warnings and for monitoring.
Gait Recognition
 Short distance: Cooperative

Finger Print Iris Face

Long distance: Uncooperative

Gait
Gait Recognition

As a biometric, gait is still available at a distance when other

biometrics are obscured or at too low resolution. Therefore, we need
gait recognition.
Applications of Gait Recognition
Gait Recognition System
Feature vector DF = {D1, D2, D3, D4, θ} of dimension 5 is used to
characterize the gait.

An Instance of feature extraction

Silhouette with different covariates
Gait Matching and Recognition
Recognition Accuracy for CASIA B Database
Although the issue denotes a relatively long-time challenge, most of the
techniques developed to handle the problem present several drawbacks
related to feature extraction and low classification rates, among other
issues.

Deep learning-based approaches recently emerged as a robust set of tools

to deal with virtually any image and computer-vision-related problem,
providing paramount results for gait recognition as well.
Deep Learning-based Gait Recognition Approaches Organized by Type of Neural
Network

Gait Recognition Based on Deep Learning: A Survey

ACM Computing Surveys, Vol. 55, No. 2, Article 34. Publication date: January 2022.
Schematic diagrams from deep learning (top) and standard (bottom) gait recognition
pipelines. As the image suggests, deep learning approaches abstract the steps
concerning data pre-processing, feature extraction, and classification in a single
architecture.
Example frames from CASIA B dataset. Each image corresponds to one of the 11
comprised angles.

Example frames from CASIA B dataset. One can notice the change in clothes and
personal objects, like backpacks.
 Example frames from
CASIA C dataset. The
images were obtained
through an infrared camera
at night and with variations
in the manner of walking.

Example frames from CASIA D

dataset. The images illustrate the
images’ acquisition process
synchronously with the footsteps data
Example of the multiview
dataset. The image presents
people walking in different
directions, from multiple
points of view
There are a few critical points about video-based models and databases
that are worth highlighting:

 The presented datasets do not introduce more than one person per
video, either for training or validation purposes.

 The environments in which they were recorded are fully controlled.

Despite the variation of angles in some sets, it is possible to notice
no variation of objects or even of colors in the background.

 Other videos were recorded with people walking on electric

treadmills, denoting even greater control of the movements and the
environment.
 Challenges

 Discriminating Features not well understood.

 Style of walking.
 Human profile.
 Coordinated movement of limbs, and torso.
 Speed of walking.
 High degree of Freedom (or variation) of movement of subjects.
 Orientation of torso, carrying condition, etc.
 View dependent.
 Frontal, fronto-parallel, back view
 Presence of multiple subjects.
 Occlusion.
Challenges
 Multi-view depth images.

 Partial gait cycle.

 To derive view independent representation.

 Fusion of multi-view information.

 Reconstruction of a complete gait cycle?

 To establish correspondences using partial gait in a different view.

 To maintain and update gallery images.

 Occlusion

 Presence of multiple persons.

Summary and Conclusion

 In gait recognition, rate of recognition will be good when both training and
testing are in same view, but if there is change in training and testing view,
recognition rate drops. This issue is to be addressed, i.e., view independent
in training and testing.

 The rate of recognition is good for normal walking, but recognition rate
drops when different covariates such as holding an object, carrying a back
pack, different clothes etc., are given as probe. More attempts must be
made to improve the accuracy of gait recognition invariant to covariates, so
that gait recognition techniques can be applied to real applications.
 Contd…

 Capturing gait information from frontal view is very difficult for gait
recognition. Some novel methods must be made attempted to focus good
accuracy in gait recognition for frontal view.

 The recent developments in gait research indicate that gait technologies still
need to mature and that limited practical applications should be expected in
the immediate future.

 At present, there is a potential for initial deployment of gait for recognition

in conjunction with other biometrics.

 Future advances in gait analysis and recognition are expected to result in

wide deployment of gait technologies not only in surveillance, but in many
other applications as well.
References
 Human identification at a distance via gait recognition: Liang Wang
Center for Research on Intelligent Perception and Computing (CRIPAC)
National Lab of Pattern Recognition (NLPR)
Institute of Automation, Chinese Academy of Sciences (CASIA)

 Nikolaos V. Boulgouris and Zhiwei X. Chi, “Human gait recognition based on

matching of body components”, Pattern Recognition 40 (2007) 1763-1770.

 Ju Han and Bir Bhanu, “Individual Recognition using Gait Energy Image”, IEEE
Trans on Pattern Analysis and Machine Intelligence, Vol. 28, no.2, February 2006.

 Sudeep Sarkar, P. Jonathon Philips, Zongyi Liu, Isidro Robledo Vega, Patrick
Grother and Kevin W. Bowyer, “The HumanID Gait Challenge Problem: Data sets,
Performance, and Analysis”, IEEE Trans. Pattern Analysis and Machine
Intelligence, Vol. 27, no.2, February 2005.
References
 Liang Wang, Tieniu Tan, Huazhong Ning and Weiming Hu, “Silhouette Analysis-
Based Gait Recognition for Human Identification”, IEEE Trans. Pattern Analysis
and Machine Intelligence, Vol. 25, no.12, December 2003.

 Mridul Ghosh and Debotosh Bhattacharjee, “Human Identification by using corner

points”, International journal of Graphics and Signal processing, 2012, 2, 30-36.

 Yumi Iwashita, Adrian Stoica and Ryo Kurazume, “Gait identification using
shadow biometrics”, Pattern Recognition Letters 33 (2012) 2148-2155.
Thank You All
 Multimodal Biometric Systems:
Research Progress
Dr. Sharmila Kumari, M
Professor and Head,
Department of Computer Science and Engineering
P.A. College of Engineering, Mangalore.
[email protected]

AICTE Sponsored ATAL FDP on “Research Avenues in Biometrics Using Low Cost Devices”
Department of Computer Applications
JSS Science and Technology University, Mysuru.
Introduction
IBM Power Systems

• Automated method for recognizing individuals based on measurable biological and

behavioral characteristics.

2
Different biometric traits
IBM Power Systems
Biometric Recognition Systems
IBM Power Systems
Why Biometrics
IBM Power Systems
Applications
IBM Power Systems

• Missing Persons • Computer login • Passport control

• Corpse identification • Electronic Payment • Border control
• ATMs • Access Control
• Criminal investigations
• Record Protection
Most Popular Traits
IBM Power Systems

• Legacy database • Legacy database • Gaining popularity

• 1:N search • 1:N search • 1:N search
• NIST evaluation • NIST evaluation • NIST evaluation
Aadhar IBM Power Systems

“Issue a unique identification number (UID) that can be verified and

authenticated in an online, cost-effective manner, and that is robust to
eliminate duplicate and fake identities.”

Basic demographic data and

1568 3647 4958 biometrics stored centrally
Name
Parents
Gender UID = 1568 3647 4958
DoB
10 fingerprints, 2 iris & face image
PoB
Address

Central UID database

The UIDAI has issued a total of 122.90 crore Aadhaar cards as on 30 November 2018.

•https://uidai.gov.in/
Challenges IBM Power Systems

• Uncooperative users: Face, Iris, capturing challenges…..

• Unconstrained imaging conditions: Occlusion, Illumination
Variations….
• Image quality: Noisy, sensing/Capturing device….
• Intra-class variation
• Inter-class similarity
• Attacks on biometric systems
• User privacy

Salient representation and robust matching

Intra-class Variability
IBM Power Systems

Same face under different expression, pose, illumination

Intra-class variability- Aging…. IBM Power Systems
Inter-class similarity
IBM Power Systems

Identical twins Celebrities

Unconstrained and aging problem
IBM Power Systems
Fake or Spoof Biometric Input IBM Power Systems
From Constrained to Unconstrained
IBM Power Systems

Unconstrained
54% TAR @ 72% Rank-1 66.8% TAR @
FAR=0.1% accuracy FAR=10%

MBGC FVC2004 CASIA.v4-distance LFW NIST SD27 UBIRIS.v2

Imaging Conditions

100% TAR 99.4% TAR 97.8% TAR

@ FAR=0.1% @ FAR=0.01% @ FAR=0.01%
Constrained

FRGC, Exp. 1 FpVTE 2003 IREX III FERET User distorted image IIITD alcoholic iris
FVC2006

Users
Cooperative Uncoopera3t2ive
 IBM Power Systems

Face Biometric
Research Developments
Face Biometric
IBM Power Systems
PCA based face recognition
IBM Power Systems
Local Binary Pattern for face recognition IBM Power Systems
Local Binary Pattern based face recognition
IBM Power Systems
Local Descriptor based face recognition - David Lowe ,1999
IBM Power Systems
Deep Learning based Face recognition
Face Datasets
IBM Power Systems
Mobio Dataset
IBM Power Systems

The MOBIO database consists of bi-modal (audio and video) data taken from 150 people.
The Idiap Research Institute,Switzerland- is an independent, nonprofit research foundation in the Swiss Alps.
IJB-A (IARPA Janus Benchmark)Dataset
IBM Power Systems
 IBM Power Systems

Fingerprint
Research Developments
Major milestones in the history of fingerprint recognition
IBM Power Systems
Significant features
IBM Power Systems
Taxonomy of fingerprint biometric system
IBM Power Systems
 IBM Power Systems

Iris Biometric
Research Developments
Why Iris Biometric
IBM Power Systems
Historical milestones in the development of iris recognition system
IBM Power Systems
Iris characteristics
IBM Power Systems
Framework for Iris recognition
IBM Power Systems
History of Iris recognition methods
IBM Power Systems
Convolution Neural Network based Model for Iris recognition
IBM Power Systems
Contributors for iris datasets
IBM Power Systems
Challenges in Iris biometric
IBM Power Systems
Challenges in Iris biometric
IBM Power Systems
Spoofing and Liveness datasets
IBM Power Systems
Multimodal Biometric System
IBM Power Systems
Integration of multiple modalities
IBM Power Systems
Score normalization in multimodal biometric systems
IBM Power Systems
IBM Power Systems
Enhanced multimodal biometric recognition approach
IBM Power Systems
Multimodal biometric recognition systems using deep learning based on the finger vein
and finger knuckle print fusion (IET Image Processing, 2020, Vol. 14 Iss. 15, pp. 3859-3868)
IBM Power Systems
Multimodal recognition system based on feature level fusion

FKP- Hong Kong Polytechnic database, FV- Database aquired by Shandong University
Multimodal recognition system based on score level fusion
IBM Power Systems
Issues in Designing Multimodal Biometric System
IBM Power Systems

• Challenges
• Choice for combination of biometric traits.
• Level of fusion for multiple modalities.
• Feature representation and matching techniques.
• Demand based compromise between authentication performance, computations and cost.
• Storage space required for templates

• Advantages
• Improves the overall accuracy
• These systems are more reliable
• It tackles the problem of non-universality
• Protects against spoof attack
• Multimodal authentication provides more level of authentication.
THANK YOU
 ATA L F D P R e s e a r c h A v e n u e s i n B i o m e t r i c s u s i n g
Low Cost Devices

NATIONAL EDUCATION POLICY-2020

and
Implementational Challenges for
Higher Education
Date: 14th December 2022

1
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

People Involved
COMPOSITION OF THE
COMMITTEE

❑ K. Kasturirangan (Chairman)
❑ Vasudha Kamath
❑ Manjul Bhargava
❑ Ram Shankar Kureel
❑ T.V. Kattimani
❑ Krishna Mohan Tripathy
❑ Mazhar Asif
❑ M.K. Sridhar

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Process

❑Discussions with educators, researchers, policy

makers, sector experts, industry, academies, community
groups, engaged citizens and more…
❑Consultations with 70 organizations, institutions,
associations and 216 eminent individuals (from July
2017 onwards)
❑ Peer Reviewers: Jayaprakash Narayan, P. Rama Rao,
J.S. Rajput, Vijay Kelkar, Aniruddha Deshpande, Dinesh
Singh, Mohandas Pai.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Relevant work and Knowledge base

from the past

❑ Policies of 1968, 1986 and 1992

❑ TSR Subramanian Committee Report

❑ MHRD’s Inputs for Draft National Education
Policy
❑ Feedback from ground level consultations since
2015

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

NEP-2020: Importance
Is very important as this New Education
Policy is coming out after 25 years of
globalisation(1992)in India.
The Policy (NEP-19) is framed after the
introduction of Internet which has done
tremendous effect on Indian Education
System.
Heavy involvement of Private players in
Indian Education.
JSSSTU, Mysore Dean - Academics
9 December 2022 22
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

UNIVERSITY EDUCATION SYSTEM

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Objectives: NEP-2020

❑ Revamp the higher education system

❑ By Creating world class multidisciplinary
higher education institutions across the
country
❑ To Increase GER to at least 50% by 2035.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices
MAJOR PROBLEMS CURRENTLY FACED BY THE
HIGHER EDUCATION SYSTEM IN INDIA

▪ A severely fragmented higher educational system;

▪ Less emphasis on the development of cognitive
skills and learning outcomes;
▪ A rigid separation of disciplines, with early
specialization and streaming of students into
narrow areas of study;

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices
MAJOR PROBLEMS CURRENTLY FACED BY THE
HIGHER EDUCATION SYSTEM IN INDIA
▪ Limited access particularly in socio-economically
disadvantaged areas, with few HEIs that teach in
local languages .
▪ Limited teacher and institutional autonomy;
▪ Inadequate mechanisms for merit-based career
management and progression of faculty and
institutional leaders;

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices
MAJOR PROBLEMS CURRENTLY FACED BY THE
HIGHER EDUCATION SYSTEM IN INDIA

▪ Lesser emphasis on research at most universities

and colleges, and lack of competitive peer-
reviewed research funding across disciplines;
▪ Suboptimal governance and leadership of HEIs;
▪ An ineffective regulatory system; and (j) large
affiliating universities resulting in low standards
of undergraduate education.

JSSSTU, Mysore
ATA L F D P R e s e a r c h A v e n u e s i n B i o m e t r i c s u s i n g
Low Cost Devices

NEP-2020
Key Changes Proposed

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

NEP-2020: Key Changes Proposed

▪ Moving towards a higher educational system consisting

of large, multidisciplinary universities and colleges, with
at least one in every district, and with more HEls across
India that offer medium of instruction in local/Indian
languages;
▪ Moving towards a more multidisciplinary
undergraduate education;
▪ Moving towards faculty and institutional autonomy;

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

NEP-2020: Key Changes Proposed

▪ Revamping curriculum, pedagogy, assessment, and
student support for enhanced student experiences;
▪ Reaffirming the integrity of faculty and institutional
leadership positions through merit-appointments and
career progression based on teaching, research, and
service;
▪ Establishment of a National Research Foundation to fund
outstanding peer-reviewed research and to actively seed
research in universities and colleges;

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

NEP-2020: Key Changes Proposed

▪ Governance of HEIs by high qualified independent boards
having academic and administrative autonomy;
▪ "Light but tight" regulation by a single regulator for higher
education;
▪ Increased access, equity, and inclusion through a range of
measures, including greater opportunities for outstanding
public education; scholarships by private/philanthropic
universities for disadvantaged and underprivileged students;
online education, and Open Distance Learning (ODL); and all
infrastructure and learning materials accessible and available to
learners with disabilities.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

NATIONAL EDUCATION POLICY -2020

TOWARDS RE-ENERGISING OF
HIGHER EDUCATION

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

NATIONAL EDUCATION: KEY

FACTORS
The New Education Policy(NEP-2019) envisions a
complete overhaul and re-energising of the
higher education system in order to overcome the
eight challenges and thereby deliver high quality
higher education, with equity and inclusion, to all
young people who aspire to it.

JSSSTU, Mysore
ATA L F D P R e s e a r c h A v e n u e s i n B i o m e t r i c s u s i n g
Low Cost Devices

#1
FRAGMENTATION OF THE
HIGHER
EDUCATION SYSTEM

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Fragmentation of the Higher

Education System
India has over 800 universities and approximately
40,000 colleges, reflecting the overall severe
fragmentation and small size of HEIs currently in
the country.
Over 40% of all colleges in the country run only a
single programme, far from the multidisciplinary
style of higher education that will be required in
the 21st century. In fact, over 20% of colleges have
enrolment below 100, while only 4% of colleges
have enrolment
9 December 2022 over 3000 (AISHE 2016-17).
JSSSTU, Mysore Dean - Academics
37
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Fragmentation of the Higher

Education System
To make matters worse, thousands of the smaller
colleges hardly have any teaching faculty at all,
and there is little or no education taking place -
thus affecting severely the integrity of the higher
education system in the country.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Fragmentation of the Higher

Education System
This fragmentation of the system leads directly to
severe suboptimality(less than the standard or
quality) on various fronts:
❑ Resource utilisation
❑ The range and number of programmes and
disciplines, The range and number of faculty.
❑ The ability to carry out high-quality
multidisciplinary research.
JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Fragmentation of the Higher Education

System
Too many silos; Too much early Specialisations and
streaming of students into disciplines:

India’s higher education has rigid boundaries of

disciplines, along with a narrow view of what
constitutes education.

As already mentioned, its most harmful expression is

in the enormous number of mono-field institutions that
have been developed, most notably in the professional
9 December 2022JSSSTU, Mysore Dean - Academics
and vocational fields.
40
ATA L F D P R e s e a r c h A v e n u e s i n B i o m e t r i c s u s i n g
Low Cost Devices

II

LACK OF ACCESS,
ESPECIALLY IN
SOCIO-ECONOMICALLY
DISADVANTAGED AREAS

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices
Lack Of Access, Especially In Socio-Economically
Disadvantaged Areas
Access in higher education has significantly improved in the
past few decades, but is still not sufficient to reach all our
young citizens; equity in and quality of education still remain
a big challenge.

While the GER of higher education has risen over the last
several years, to around 25%, and notable progress has been
made, this Policy (NEP-2019)aims for GER to reach 50% by
2035, in order to fulfil the aspirations of our youth and to
form the basis for a vibrant society and economy.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

III

LACK OF
TEACHER AND
INSTITUTIONAL AUTONOMY

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Energised, Engaged and

Capable Faculty
Objective: Empowered faculty with high
competence and deep commitment, energised for
excellence in teaching and research.

The most important factor in the success of higher

education institutions is the quality and engagement
of its faculty.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Education Policies and Teachers

in INDIA

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices
Mudaliar Commission (1954) – About Teachers:

The Education Commission or Mudaliar Commission

(1954) says :
“The most important factor in the contemplated
educational reconstruction is the teacher, his personal
qualities, his educational qualifications, his professional
training and the place, he occupies in an Institution as
well as in the communication.

The reputation of an Institution and its influence on the

life of community invariably depend on the kind of
teacher working in it”.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

National Policy on Education (1968)

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

National Policy on Education (1986)

According to NPE (1986):

“The status of the teacher reflects the socio-cultural

ethos of a society. It is said that no people can rise above
the level of its teachers. The government and the
community should endeavour to create conditions
which will help motivate and inspire teachers on
constructive and creative lines.

After 34 Years ????

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

National Policy on Education (1992)

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Real Challenges for Teachers

In HEIs
(As per NEP-2020)

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Challenges for faculty motivation in HEIs

▪ First, physical infrastructure and service

conditions remain less than ideal at too many
institutions, including at many Central and State
Universities.

▪ Many institutions lack the basic facilities and

infrastructure needed for faculty (and students) to
feel comfortable in coming to work.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Challenges for faculty motivation in HEIs

▪ The service conditions of faculty members are also severely
inadequate. At the current time, there are too many faculty
members on temporary appointments, with low salaries
and/or insecurity.

▪ Ad-hoc and contractual appointments have become the

norm, compromising institutional processes and depleting
the energies and motivation of all faculty members.

▪ This is true for both public and private HEIs.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Challenges for faculty motivation in HEIs

▪ Heavy teaching loads (often as much as 36 hours a
week), with high student-teacher ratios in each class
(sometimes higher than 40), leave little time for
adequate class preparation or proper student
interaction, let alone time for research or other
university activities and service.

▪ Faculty completely lack autonomy - in course

development, on curricular matters, on pedagogical
approaches, on taking service initiatives, and on
research.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Challenges for faculty motivation in HEIs

A further challenge with faculty motivation is
that career management is too often not based on
merit, but rather on seniority, luck, or other
arbitrary factors.
There is no clear tenure track or career
progression system at most institutions that
ensures that hiring, retention, salary increase,
promotion, and vertical mobility are all based on
merit and quality of performance in teaching,
research, and service.
JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Challenges for faculty motivation in HEIs

Incentives for conducting

outstanding work are not
an inherent part of the
system, severely reducing
faculty motivation and
commitment to excellence.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Challenges for faculty motivation in HEIs

Institutional leadership:
Institutional leaders are
not trained and fostered
well in advance, or always
chosen based on merit; in
fact, in too many cases,
institutional leadership is
chosen based on totally
corrupt practices.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Challenges for faculty motivation in HEIs

Finally, the institutional leadership

system is broken. Institutional leaders are
not trained and fostered well in advance,
or always chosen based on merit; in fact,
in too many cases, institutional leadership
is chosen based on totally corrupt
practices.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Challenges for faculty motivation in HEIs

Transition between institutional leaders is too often not

smooth, with reports of leadership vacancies for several
months or more.

As institutional leadership, by definition, must take the

lead in creating a merit-based culture of excellence and
high performance at each institution, the broken system
of leadership can have a severely demotivating effect on
both faculty and students.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

TEACHER AUTONOMY

The concept of refers to

the professional independence of teachers
in Colleges / Institutions , especially the degree
to which they can make autonomous decisions
about what they teach to students and how they
teach it. ...

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

TEACHER AUTONOMY
Teacher autonomy is essential for ensuring a learning
environment that addresses children's diverse needs.

As much as the learner requires space, freedom,

flexibility and respect, the teacher also requires the same.
There is need to encourage an atmosphere that facilitates
collaborative efforts among teachers.

Teacher autonomy is driven by a need for personal and

professional improvement, so that an autonomous
teacher may seek out opportunities over the course of his
or her career to develop further.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

PROBLEM ASSOCIATED
WITH TEACHERS

SILO EFFECT

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

SILO EFFECT
Many Professors / Faculty members suffer from
what has been called “the silo effect.”

This occurs when they become isolated in their

own little part of their academic neighbourhood
and consequently experience minimal
interaction with colleagues.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

SILO EFFECT
Such isolation results from the pressures of class
preparation, grading, advising, committee work, service
to the institution, continuing education, writing and
publishing, participation in conventions, speaking
engagements, additional teaching duties for other
institutions, community service, and other services.

And of course, one must also have some personal time

and family time.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

“SILO Effect”
To examine your own experience of the “silo effect,” reflect on these
questions:
❑ How many different colleagues do you interact with each week?
❑ How many hours per week do you spend interacting with
colleagues? How many hours do you spending working alone?
❑ How satisfied are you with the amount of interaction that you have
with colleagues?
❑ How satisfied are you with the quality of interaction that you have
with colleagues?
❑ How does the “silo effect” affect you negatively?

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

SILO EFFECT ON AN INSTITUTION

The “silo effect” is also detrimental to the

well-being of an institution of higher
education (IHE).

People are not as effective or productive

when they work alone as they are when they
collaborate with others.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices
SILO EFFECT-ELIMINATION
So how does the professor break through the silo
and form networks of relationships across
campus and across departments?

One solution is to participate in a faculty learning

community (FLC-JNANAVARDHAN).
FLCs provide an intentional strategy for breaking
down the barriers between faculty members and
encouraging interaction that will help them
improve their effectiveness as educators while
also meeting some of their personal needs.
JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Lack of teacher and institutional autonomy

The lack of teacher autonomy has led to a severe lack of
faculty motivation and scope for innovation.
In order for faculty members and institutional leaders to
innovate and explore in their teaching, research, and
service, they must have the individual autonomy that
allows them to do so.
In particular, the system of affiliated colleges which are
required to follow a central syllabus, curriculum, pedagogy,
and textbook makes it very difficult to provide teachers
with such autonomy.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Lack of teacher and institutional

autonomy

Most institutions and institutional leaders are unable

to take bold and innovative steps to enhance their
educational offerings or research related and
community outreach programmes, because they too
do not have the academic, administrative, or financial
autonomy to do so.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Lack of teacher
and institutional autonomy
A final challenge in recent years is that the very
word ‘autonomy’ has come to mean ‘reduction of
public funding’,

‘Autonomy’ must mean the freedom to innovate,

to compete, to cooperate, to govern more locally,
to optimise resources given one’s direct local
knowledge of circumstances and opportunities, to
break silos, and to excel.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

IV

INADEQUATE MECHANISMS FOR

CAREER MANAGEMENT
AND
PROGRESSION OF FACULTY MEMBERS
AND
INSTITUTIONAL LEADERS

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Reason for lack of novel initiatives from faculty and

institutional leaders is:
The lack of a suitable structure for career management of
faculty and of institutional leaders.
The system of selection, tenure, promotion, salary increases
and other recognition and vertical mobility of faculty and
institutional leaders is, at the current time, not based on merit
but tends to be either seniority based or
arbitrary.
This has had the negative effect of severely disincentivising
quality and innovation at all levels.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

THE LACK OF RESEARCH AT MOST UNIVERSITIES

AND COLLEGES, AND THE LACK OF
TRANSPARENT
AND
COMPETITIVE PEER REVIEWED RESEARCH
FUNDING ACROSS DISCIPLINES:

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

THE LACK OF RESEARCH AT MOST UNIVERSITIES

The separation in higher education between

teaching institutions and research institutions
post-independence has caused much harm, as
most universities and colleges in the country
today conduct very little research.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

THE LACK OF RESEARCH AT MOST UNIVERSITIES

This has resulted in :
First, so many members of the academic community of the
country not conducting (and not incentivised to conduct)
scholarly research is an enormous lost opportunity for
research and innovation in the country.
Secondly, on the education side, it is difficult to have
outstanding higher education and teaching in an
environment where knowledge creation is not taking place;
indeed, how can students be taught to innovate in a
location where innovation is not on the agenda.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

THE LACK OF RESEARCH AT MOST UNIVERSITIES

At the current time, there is no mechanism to seed or mentor

research at universities and colleges where research is in a
nascent stage particularly at State Universities where 93% of
all students in higher education are enrolled.

In addition, very little funding is available for novel research

ideas and proposals at most HEIs, especially for research in
important multidisciplinary and cross-disciplinary areas (e.g.
relating to clean water, renewable energy, education and
pedagogy, health, etc.)

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

A regulatory system allowing

fake colleges to thrive
while
constraining
excellent, innovative institutions

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

A regulatory system allowing

fake colleges
There are many fake colleges in existence that run with
impunity, while excellent colleges and universities feel
constrained academically, administratively, and financially.

Regulation has been too heavy handed for decades. It has

been the key contributor to the diffused sense of
autonomy and accountability in the system. Too much has
been attempted to be regulated with too little effect.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

The mechanistic and disempowering regulatory system

has been rife with basic problems, such as a
concentration of power within a few bodies, conflicts of
interest among these bodies, and a resulting lack of
accountability.
This regulatory culture has unfortunately stifled
innovation and creativity, and unfortunately instead has
fostered mediocrity and graft.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Ill-treated Private Institutions

Private HEIs have not been treated on an equal footing with
public institutions.
On the one hand this approach has discouraged public-
spirited philanthropic HEIs, while on the other hand it has
been unable to stop commercialisation of education.

JSSSTU, Mysore
 ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

CONCLUSION

NEP-2020 has recommended very important points with

reference to Empowering and Energising the Teachers.
Most of these recommendations are already available in
Education Policies of 1954 , 1968 and so on….

Hoping that, the “ Teacher” will be energised and

empowered in true spirit by the stake holders during this
NEP-2020.

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

Conclusion:

•Need of the hour

•Reformation internally and externally
•Incorporating the facts to fill the gap
•Acknowledging the challenges
•Implementation of NEP 2020

JSSSTU, Mysore
Acknowledgements
My heartfelt thanks to my Good Friend
Dr. G Thippeswamy
Professor and Dean(Academics)
BMSIT Bangalore.

https://www.education.gov.in/sites/upload_file
s/mhrd/files/NEP_Final_English_0.pdf

https://pib.gov.in/PressReleaseIframePage.as
px?PRID=1847066
ATAL FDP Research Avenues in Biometrics using
Low Cost Devices

JSSSTU, Mysore
 Research Scope for Developing
Efficient Biometric applications in the Compressed Domain

By
Dr. Mohammed Javed
Assistant Professor, Department of IT
Indian Institute of Information Technology (IIIT-Allahabad)
Email:[email protected]

TWO WEEK AICTE sponsored ATAL FDP ON

“RESEARCH AVENUES IN BIOMETRICS USING LOW COST DEVICES”
By Department of Computer Applications, JSS Science & Technology University, Mysore

December 10, 2022

Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
Popular Biometrics in Use and Challenges Ahead

1) Finger Print
2) Voice Recognition
3) Iris Recognition
4) Facial Recognition
5) Handwriting Recognition
• Banking and Judicial
• Doesn’t need advance devices
Image Courtesy: https://www.bayometric.com/biometric-system-architecture/

Future Challenges: Physical Spoofing, DeepFakes, Civil and Legal Accountability

Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
Sources of Big Data and Motivation for Compression

Data Compression

1. Low Bandwidth while Data Transmission

2. Less Space while Data Archiving

Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
Biometric Samples as Document Images

Courtesy: https://richtechit.wordpress.com/2014/03/28/cheque-printing-software/

Courtesy: https://sbnri.com/blog/passport/passport-for-minor-in-india
Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
Conventional Processing in Uncompressed Format

Decompression
Compressed Processing and
Data Analytics

Re-Compression
Big Data Compressed
(Large Volume Data) Data
(Small Size Data)
decompression


 Additional Computing Resources (Time and Buffer Space)

 During Processing Objective of Compression is not Achieved

 Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
Can the Processing also be Faster in Compressed Format

If Texts, Images, Audios and Videos are all stored and transmitted in the Compressed form
Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
Working in Compressed Domain

• Conventional • Conventional
Or Or
Handcrafted Handcrafted

Pixel Compressed
Domain Domain

Pixel Compressed
Domain Domain

• Deep Learning • Deep Learning

 Dr. Mohammed Javed
Assistant Professor, Department of IT Compressed Document of a Printed Text
IIIT-Allahabad

Some Challenges like

Courtesy: https://richtechit.wordpress.com/2014/03/28/cheque-printing-software/  Text Line Segmentation
 Entropy Computation
 Equivalence Detection

Courtesy: https://sbnri.com/blog/passport/passport-for-minor-in-india
 Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad

Slide Courtesy: Camelia Floria, 19th Summer School on Image Processing, 2011
 Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad

Slide Courtesy: Camelia Floria, 19th Summer School on Image Processing, 2011
 Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad

Published Research works Directly in JPEG Compressed Domain

Bulla Rajesh, Mohammed Javed, P. Nagabhushan,"Segmentation of Printed Text-Lines Directly in

JPEG Compressed Document Images", Published in Proceedings of IEEE GCCE2019, Pages 1091-
1092, October 15-18, 2019, Osaka, Japan

Bulla Rajesh, Mohammed Javed, P. Nagabhushan, “ Automatic Tracing and Extraction of Text-Line
and Word Segments Directly in JPEG Compressed Document Images”, Published in IET Image
Processing, Vol. 14(9), Pages 1909 – 1919, 2020

Bulla Rajesh, Mohammed Javed, P. Nagabhushan,"Segmentation of Text-Lines and Words from JPEG
Compressed Printed Text Documents Using DCT Coefficients”, Published in IEEE Data Compression
Conference (DCC2020), Page 389, March 24-27, 2020, UTAH, USA (A* conference by CORE Ranking)
Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad

Towards
Deep Learning based Intelligent Processing
of Documents in Compressed Domain
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Feeding of JPEG compressed stream to CNN

The process for the preparation of

the JPEG compressed input
through partial decompression,
and feeding the same to the deep
learning model

1. L. Gueguen, A. Sergeev, B. Kadlec, R. Liu, and J. Yosinski, “Faster neural networks straight from jpeg,” NeurIPS, 2018.
2. Benjamin Deguerre, Clément Chatelain, Gilles Gasso, "Fast object detection in compressed JPEG Images”, ITSC, 2019.
3. Ehrlich and L. Davis, “Deep Residual Learning in the JPEG Transform Domain”, ICCV, 2019.
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Feeding of JPEG compressed stream to CNN

Faster Networks
 Explained the procedure how to feed JPEG compressed visual data to different
deep learning models trained on other existing data
 Reported faster computations keeping the recognition performance same.
 Did not analyzed the textual data
 Did not explore a novel deep learning model for compressed data.

Fast object detection in Compressed domain

 A modified SSD architecture is used to detect the objects in the outdoor images.
 Analyzed data by introducing the 8x8 filter setting.
 Reported 2 time faster results.
 Did not analyzed the textual data

1. L. Gueguen, A. Sergeev, B. Kadlec, R. Liu, and J. Yosinski, “Faster neural networks straight from jpeg,” NeurIPS, 2018.
2. Benjamin Deguerre, Clément Chatelain, Gilles Gasso, "Fast object detection in compressed JPEG Images”, ITSC, 2019.
3. Ehrlich and L. Davis, “Deep Residual Learning in the JPEG Transform Domain”, ICCV, 2019.
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

JPEG compressed Input preparation

 Converting the color image having (R,G,B) channels to Y, Cb, Cr channels.

 During this process there would be a variation in the resolution between Y and Cb, Cr components.
 Applying the DCT operation on each channel separately.
 Concatenation of these three channels to feed to the DCT-CompCNN model.

B Rajesh, M Javed et al, “DCT-CompCNN: A novel image classification network using jpeg compressed DCT coefficients”, IEEE CICT 2019
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Proposed DCT-CompCNN Model for JPEG Compressed Stream extracted from JPEG Images

B Rajesh, M Javed et al, “DCT-CompCNN: A novel image classification network using jpeg compressed DCT coefficients”, IEEE CICT 2019
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

 The experimentations were done on ResNet-50 and DCT-CompCNN models using quantized and
unquantized coefficients.

 The observation is that there is less difference in accuracy in case of quantized JPEG images and un-
quantified JPEG images, which shows that low frequency regions are more responsible for
learning the features from a compressed image.

 We also analyzed that when we Down-sampled the luma channel during transformation, we achieved less
inference speed, but lesser accuracy, which is vice versa in case of Up-sampled chroma channel.

 These models are trained on NVIDIA 1x Tesla K80 and have Intel Xeon 2.3 GHz Processor (provided by
Google Collab).

 The ResNet-50 which have around 23 million parameters with a training time of roughly 1022 seconds for
each epoch.

 In our proposed DCT-CompCNN model which was trained on the CIFAR-10 and MNIST datasets in which
we have around 1.8 million parameters with a training time of roughly 66 and 40 seconds per one epoch
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Classification Result And Inference Speed(in Sec) For Resnet-50 Model Tested On Dogs Vs Cats Dataset

Type of Operation Accuracy (%) Inference

Speed
Quantized (Down- Sampling) 81.63 880 ± 5
Quantized (Up- Sampling) 83.56 1072 ± 5
Unquantized (Down- Sampling) 82.43 874 ± 5
Unquantized (Up- Sampling) 84.78 1084 ± 5

Classification Result And Inference Speed(in Sec) For Proposed DCT-CompCNN Model Tested On CIFAR 10
And MNIST Datasets

Type of Operation Accuracy (%) Inference Speed

Quantized (Down- Sampling) 81.63 880 ± 5

Quantized (Up- Sampling) 83.56 1072 ± 5
Unquantized (Down- Sampling) 82.43 874 ± 5
Quantized (Up- Sampling) 84.78 1084 ± 5
Quantized (MNIST) 98.11 40 ± 3
Unquantized (MNIST) 99.52 38 ± 3
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

 DCT-CompCNN model has been proposed for classification of the direct compressed data.

 Tested with all types of JPEG compressed input streams including quantized and
Unquantized DCT
 Coefficients with both down sampling and up sampling of Y and Cb, Cr channels.
 Further, we tested on Gray scale images taking standard dataset (MNIST).

 This is an attempt to make CNN model for direct compressed data possible.

 This would be a milestone step for changing all conventional / pixel domain based models
into compressed domain.
 Compressed domain has been proved to be efficient in terms of execution time and space.
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Holistic Handwritten Word Recognition in the JPEG Compressed Domain

 Holistic word recognition techniques are often used to analyze handwritten documents where the
vocabulary is known.
 They are computationally less expensive than the methods employing character segmentation and
recognition

The encoding of (a) handwritten word image during 8x8 DCT transformation, (b) DCT encoded word image.

B Rajesh, P Jain, M Javed, “HH-CompWordNet: Holistic Handwritten Word Recognition in the Compressed Domain”, IEEE, DCC2021
Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Holistic Handwritten Word Recognition in the JPEG Compressed Domain

Sample Bangla handwritten word images in both pixel and DCT compressed representations

B Rajesh, P Jain, M Javed, “HH-CompWordNet: Holistic Handwritten Word Recognition in the Compressed Domain”, IEEE, DCC2021
Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

 Proposed HH-CompWordNet Model

 Holistic compressed word recognition model in JPEG compressed domain

B Rajesh, P Jain, M Javed, “HH-CompWordNet: Holistic Handwritten Word Recognition in the Compressed Domain”, IEEE, DCC2021
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Proposed HH-CompWordNet Deep Learning Architecture

B Rajesh, P Jain, M Javed, “HH-CompWordNet: Holistic Handwritten Word Recognition in the Compressed Domain”, IEEE, DCC2021
Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Visualization of output activation at each convolution layer for the HH-CompWordNet

Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

HH-CompWordNet with Uncompressed domain models

HH-CompWordNet with Uncompressed Domain Deep Learning models

 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Semantic Segmentation

Image Courtesy: https://keymakr.com/semantic-segmentation.php

 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Semantic/Layout segmentation in the News Papers

 It is a challenging task because of varying news layouts, diverse styles and shapes of news
segments, different languages and fonts, dynamic embedding of images and advertisements
based on the regional flavor.
Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

 Need to improve the model to

locate the object boundaries
smoothly
 Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Results of the two networks on the JPEG compressed version of the Manual dataset along with
Base line result

Results of two networks on the JPEG compressed version of the Russian dataset along with Base line
result
Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad

Experimental results of two networks on the JPEG compressed (4x4) version of Manual dataset
Dr. Mohammed Javed
Assistant Professor, Department of IT Deep Learning based Compressed Domain Processing
IIIT-Allahabad
Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
Face Detection in Compressed Domain

Reference: Fonseca et. al, Face detection in the compressed domain, ICIP2004
Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
Face Detection in Compressed Domain in JPEG and JPEG 2000

Reference: Delac et. al, Face recognition in JPEG and JPEG2000 compressed domain, Image and Vision Computing, 2009
Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
Face Recognition in the Compressed Domain

Reference: Moin et. al, Face recognition in JPEG compressed domain: a novel coefficient selection approach, In Signal, Image and
Video Processing Journal, 2015
Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad
ECG Based Biometric User Identification in Compressed Domain

Compressive sensing (CS) technology that combines both sampling and compression has been widely
applied to reduce the power of data acquisition and transmission

Reference: Chou et. al, Compressed-Domain ECG-Based Biometric User Identification Using Compressive Analysis, Sensors, 2020
Dr. Mohammed Javed
Assistant Professor, Department of IT
IIIT-Allahabad

Thank You
Recent trends in
Research
Dr. Sahana D Gowda
Registrar
Research
Analysis / Exploration / Inquiry / Investigation / Probe / Experimentation.

Research is becoming a more complex collaborative

activity.

There is a growing focus on the real-world application,

societal impact and commercial return of academic
research.

Technology has growing importance as a tool to enable

more complex and collaborative research across
disciplines.
 Central Idea of Research:
Trend setting VS Trend following
* Instead of discussing the apparent trends in various corners,
give a manifestation to the original thoughts on the subject.

Back to basic – scriptural guidance

* It carries enough weight to stand off its own merit.
* They are profound in their depth.
* Much of it is quoted in the test subsequently as relevant.

Natural alignment
* It may be an imperative for us to align ourselves with
nature, all the natural phenomenon.

Originality
* Ability to think independently and creativity is a great asset.
* The quality of being novel or unusual is unparalleled and
must requires in today’s social system.

“ A work should be created with a unique style and substance”

This is probably a call to all researchers, students & teachers to familiarize themselves
with the latest trends and tools associated with research methodology being
advocated in the environment.
 Need for change in Research Approach

The 20th Century The 21th Century

Information Era Knowledge Era
• Industrialization • Markets
• Transportation • Communications
• Energy, materials • Computers, networks
• Nation-states • Knowledge, bits
• Public Policy • Nationalism
Research methodology = originality x context
originality : attributable to the researcher
Context : attributable to the situation
They are directly propositional
Modern trends to be adopted in research

Strategic Exploit latest Scriptural Theory

intuition digital tools references correlation

Focus on
Leverage Race to the Freedom from
personal
technology passes format
observation
Promote good research integrity, and
contribute to improving research culture,
in your organization:

1. Small steps can make a big difference.

2. Establishing support systems can boost

morale and enhance a positive research.

3. Ensure everyone is on the same page.

4. Research culture “cafes” are an

excellent way to share best practice.

5. Organization, department and team

leaders leading by example in promoting
an excellent research culture.

6. Discuss training gaps for all team

members.

7. Embed research culture at an

institutional level.
Information technologies in Research
Power of information and computing technologies

Increased speed and scope of research

Conduct investigations
enable new forms of inquiry

Analysis of massive datasets to detect and assess the

nature of relationships that otherwise would go unseen

Digital Technologies can pose other temptations for researchers to violate the standards of scientific practices
 Research Themes

Work & Energy & Data &

Future Cities
Economy Sustainability Digital

Governance Identities & Cultural Childhood &

& Democracy Inequalities Heritage Youth

Health &
Wellbeing
Induced or Autonomous Research
Induced research is consistent
with the institution’s traditional Autonomous research is
concept of research. Such type non-directed research
of research results in conducted with an objective
incremental innovation to come with radical
Extensive innovation.
Research
Radical Research explores
Incremental research refers the view that research is
to an orientation in not a neutral tool to be
research, almost employed without bias in
all research is incremental, the search for truth.
in that it builds upon
previous work
Autonomous Research
Positive

Think BIG

Out of the Box

Radical Innovation is a transformative model that seeks to demolish and replace an existing
industry or create a whole new industry
 Creating Value Through Autonomous
Research

1 2 3 4 5
Objective is to come Be flexible and willing Identify and exploit Foster a radical mind- Seek to enter and
up with radical to take risks. opportunities with set among compete in
innovations sufficient resources researchers. international markets
and capabilities to through transfer of
launch radical actions. technology
Career Outlook for Researchers

• Career opportunities in the research field are diverse and span a variety of
industries.
• A social science researcher may focus on areas like healthcare and
unemployment, conducting interviews and surveys to collect data for
analysis.
• In a corporate environment, an operations research analyst can help his or
her organization by reviewing business processes and identifying efficiencies.
• Market research analyst may make production recommendations after
examining consumer purchasing patterns.
 New Trends and Techniques for Researchers
Predictive Analysis
Predictive analytics refers to a sophisticated form of analysis using current and historical data to forecast future
outcomes.
Although using analytics to draw predictions about the future is not a new practice, predictive analytics is at the
forefront of data analysis because of the advanced techniques involved.

Some of the tools used in this practice include machine learning, artificial intelligence, data mining, and statistical and
mathematical algorithms.

These advanced tools and models allow for the creation of more accurate and dependable future predictions of
trends, behaviors, and actions.

The valuable information generated by predictive analytics can be used by organizations to make strategic decisions
about operations and identify opportunities and risks.
For example,
The financial services sector could use this practice to forecast market trends or create credit risk reports.
Or government and law enforcement agencies may look to gather data about community crime and use that
information to develop proactive safety measures.

Researchers need to keep abreast of this cutting-edge form of analytics because of its increasing usage.
 Real-world Use Cases of Predictive Analytics
Predictive analytics models are integrated within applications and systems to identify future results. Here are 7 real-world
real use cases of predictive analytics projects:

1.Predicting buying behavior

1. One of the biggest uses of predictive analytics is predicting buying behavior in the retail industry. Companies use
the tools to learn all about their customers. Companies use advanced analytics to identify buying habits based on
previous purchase history.
2. Walmart is a great example. It used early data to understand buying behavior in certain circumstances. Small
ecommerce retailers can incorporate predictive analytics in PoS to predict customer purchase patterns. It helps to
understand customers on a deeper, more personalized level.
2.Fraud detection
1. As cybersecurity becomes a growing concern, predictive analytics examples are plenty. The most important one is
fraud detection. These models can identify anomalies in the system and detect unusual behavior to determine
threats.
2. For example, experts can feed historical data of cyberattacks and threats to the system. When the predictive
analytics algorithm identifies something similar, it will send a notification to the respective personnel. It will limit
the entry of hackers and vulnerabilities that might put the system at risk.
 Real-world Use Cases of Predictive Analytics
3. Healthcare diagnosis
1. The healthcare industry is benefitting the most from the predictive analysis module. Health data is critical to
understand the history and current illness of any patient. Predictive analytics models help in understanding the disease
by providing an accurate diagnosis based on past data.
2. With the help of certain health factors, predictive analytics help doctors reach the root cause of diseases. It gives them
timely analytics so that they can start working on the treatments at an early stage. With the help of predictive analytics
models, the spread of negative health effects can be stopped.

4. Card abandonment
1. Retailers love this application of predictive analytics. Cart abandonment is a big issue. However, based on past history,
models can predict how likely a customer is to abandon the cart.
2. For example, by feeding the data to the model of purchase made and cart abandoned, the model predicts how many
customers will abandon it. It will also provide companies with details about each customer about whether they will
purchase or abandon the cart based on the previous visits to the store.
 Real-world Use Cases of Predictive Analytics
5. Content recommendation
1. One of the most relatable and visible predictive analytics examples is content recommendation. Through algorithms and
models, entertainment companies can predict what users want to watch based on their history.
2. When you ask, “what companies use predictive analytics?” The most relevant answer is Netflix. The entertainment
company utilizes predictive algorithms to recommend content to users based on genre, keywords, ratings, and more. The
intelligent system uses highly advanced analytics for predicting the user’s behavior.
6. Equipment maintenance
1. Predictive analytics models are significant for manufacturing, healthcare, and other businesses that require scheduled
equipment maintenance. An accidental equipment breakdown can risk human lives and lead to major losses for the
company as well.
2. For example, by integrating IoT with equipment, manufacturing units can feed data to understand when the equipment
would require maintenance. In this way, the machinery would alert the personnel and the maintenance can be done to
avoid unscheduled and accidental breakdowns.
7. Virtual assistants
1. Combined with the power of deep learning, predictive analytics works wonders when utilized with virtual assistants. Siri,
Ok Google, and Alexa are real-world use cases of predictive analytics projects. These virtual assistants learn from the
user’s behavior and then deliver accurate results.
2. Companies also use virtual assistants that act as chatbots. It improves the customer experience as these bots learn from
interactions and predict what the customer’s response would be. They are self-learning and enable companies to better
manage customers without hiring a lot of support staff.
Types of Predictive Analytics models and Algorithms
Classification model
Classification models play the basic role of sorting data and putting it in different slots. It is very vital for predictive analytics
models and algorithm as the whole prediction process begins from there. Classification models put data in categories based on
what is abstracted from historical data. Generally, this method of predictive analytics modelling is good at answering ‘yes’ or
‘no’ questions based on broad analysis. The classification model can answer basic questions like ‘Will this customer shop here
again?’ ‘Will this application pass the loan test?’ and ‘Is it a fraudulent transaction?’ Besides, when new big data comes in, the
classification models redefine themselves to house them and provide revised answers.

Forecast Models
The forecast model is very important in predictive analytics modelling, as it handles metric value prediction by estimating the
values of new data based on learnings from historical data. It is very vital because of its acceptance of both historical and current
data. Forecast models generate numerical values when there is none to be found and enhance business outcomes. It can multi-
task and also leverage answers for diverse industries without discrimination. For example, sales departments use a forecast
model to predict the number of purchase customers will make and on the other hand, it can anticipate the number of products
a manufacturing unit can produce per day.
Types of Predictive Analytics models and Algorithms
Outliner Model
When it comes to analyzing data, there is always some trouble with big data that doesn’t fit in the parameters. Outliner models
are used to counter those challenges. It analyzes abnormal or outlying data points and delivers accurate answers. Generally, the
outliner model is used to predict abnormal activities and transactions.

Clustering Model
The clustering model in predictive analytics modelling is used to target customers based on their preference. Cluster models sort
data into separate, nested smart groups based on similar attributes. Later, the company uses it to sort and promote relative
products to the customers who are most likely to buy them. Besides, the cluster model method is seen as a time-effective way of
using big data as it sorts and delivers the answers quickly. They segregate customers into smaller groups based on their
preferences and devise promotion methods specifically for each group.

Time series Model

While all the above-mentioned types focus on historical data and its outcomes, the time series model encounters the challenge
of customer preference over a period of time. Consumers have the evolving mindset when it comes to their choice of purchase.
Only the time series model gets to address those concerns in predictive analytics modelling. The time series model focuses on
data where time is the input parameter and works on different data points to develop a numerical metric to predict trends.
Six of the
1. H2O Driverless AI.
most popular predictive analytics tools.
A relative newcomer to predictive analytics, H2O gained traction with a popular open-source offering.
automating the machine learning workflow, which includes automatic training and tuning of many models within a user-
specified time limit.

2. IBM Watson Studio.

Empowers Data Scientists & Developers to Optimize Decisions.

3. Microsoft Azure Machine Learning.

Azure Machine Learning is a cloud service for accelerating and managing the machine learning project
lifecycle. Machine learning professionals, data scientists, and engineers can use it in their day-to-day workflows: Train
and deploy models and manage MLOps.

4. RapidMiner Studio.
RapidMiner is an awesome visual workflow designer. The way they present visually is so unique. It helps in
speeding and automating the creation of visual models

5. SAP Predictive Analytics.

SAP Predictive Analytics is a statistical analysis and data mining solution that enables you to build
predictive models to discover hidden insights.
6. SAS
SAS is the leader in analytics. Through innovative Analytics, Artificial Intelligence and Data Management
software and services
 Research Perspective

Societal Impact

State of the Art

Social connect

Conclusion Depth analysis

Team building

Publication

Patenting

Copywrite
Thank You