Multimedia Tools and Applications

https://doi.org/10.1007/s11042-019-08123-w

Multimodal biometric cryptosystem for human

authentication using fingerprint and ear

Padira S. V. V. N. Chanukya 1 & T. K. Thivakaran 2

Received: 21 April 2018 / Revised: 8 June 2019 / Accepted: 13 August 2019

# Springer Science+Business Media, LLC, part of Springer Nature 2019

Abstract
The multimodal biometrics is mainly used for the purpose of person certification and proof. Lot of
biometrics is used for human authentication. In which ear and fingerprint are efficient one. There
are three vital phases involved in the biometric detection which include the Preprocessing, Feature
extraction and the classification. Initially, preprocessing is done with the help of median filter which
lends a helping hand to the task of cropping the image for choosing the position. Then, from the
preprocessed Finger print and ear image texture and shape features are extracted. In the long run,
the extracted features are integrated. The integrated features, in turn, are proficiently classified by
means of the optimal neural network (ONN). Here, the NN weights are optimally, selected with the
help of firefly algorithm (FF). The biometric image is classified into fingerprint and ear if the
identical person images are amassed in one group and the uneven images are stored in a different
group. The performance of the proposed approach is analyzed in terms of evaluation metrics.

Keywords Preprocessing . Classification . Median filter . Firefly . Mismatched

1 Introduction

Biometric is an emerging technology which utilizes vein scan, facial thermogram, DNA
matching, blood pulse, ear shape, gait recognition etc. in security systems for establishing
the identity of a person. Much work is still needed for design of convenient, secure and
privacy-friendly systems. The word biometric is derived from the ancient Greek words “bios”
meaning life and “metron” meaning measure [15]. So, the meaning of biometric is life
measurement [23]. With the intention of perking up the proficiency of human detection,

* Padira S. V. V. N. Chanukya
[email protected]

1
Department of Electronics and Communication Engineering, Meenakshi Academy of Higher
Education and Research (MAHER University), K. K. Nagar, Chennai 600078 TamilNadu, India
2
Department of Computer Science and Engineering, Presedency University, yelahanka,
Bengaluru 560064, India
 Multimedia Tools and Applications

biometric investigators are leaving no stone unturned in investigating by way of exploiting the
supplementary attributes like the scars, marks, tattoos, and height and body shape in juxtapo-
sition with the basic features such as the face [21]. A traditional biometric system habitually
functions either in the enrollment mode or in the verification mode. In enrollment mode, the
system records the details of a user by amassing his/her biometric data as the reference data
known as the template in the database. The corresponding reference data is thereafter
contrasted with the query biometric data furnished by the user who claims identity in the
verification mode [4]. In this regard, hand vein based verification has surfaced as a highly
potential segment in the biometric community [19].
The problem of consistent authentication has assumed zooming significance in today’s
world. Various corporations, businesses and individuals habitually intend to limit the access to
logical or physical resources to only those who are endowed with the pertinent privileges. A
well-known technique for authentication is the effective employment of the biometric data.
However, there arises the ambiguity thanks to the fact that the deficiency of distinctiveness in
biometrics has resulted in the initiation of a host of endeavors into the multimodal biometrics
[16]. In the application of biometrics authentication (BA) techniques, the biometric data
habitually illustrates three distinct attributes such as the Titanic number of individuals,
miniature sample size and superior dimensionality. One of the crucial investigation hurdles
of the BA is the solitary sample biometrics detection challenge [28]. The speaker identification
mechanisms dependent on the voice or lip movements have been extensively perceived with
open eye-brows. Even though the unimodal biometric systems have been designed and
renovated, they are plagued by several hassles such as the noise on voice and lighting [11].
The external physical features like the fingerprint, face and iris have assumed zooming
significance these days. The Fingerprint and iris are normally deemed to facilitate further
precise biometric detection than the face. Nevertheless, the face is effectively employed in the
surveillance environments where the fingerprint and iris capture have proved to be non-viable.
Still, the face alone has failed to emerge as precise and adaptable as required for the related
environs on account of the variations in the illumination, expression, makeup or eyeglasses.
The ear constitutes a distinctive feature of human beings. It is well-established that even the
ears of “identical twins” show a tendency to differ in certain respects. Even specialists
functioning in the crime labs are of the opinion that the human external ear characteristics
are exclusive to each and every individual and do not undergo any kind of change whatsoever
in the course of the entire life cycle of an individual. For several years, various theories have
found the light of the day and duly reflected in the intermittent literature that the shapes and
characteristics of the human ear are extensively divergent and perhaps adequately variant so as
to permit the probable discrimination between the ears of the whole human race [20]. The
unimodal biometric systems are habitually encountered by several thorny issues like the noise-
polluted data, intra-class deviations, and limited scale of freedom, non-universality, spoof
assaults, and deplorable error rates [22].
The multimodal biometrics has, of late, emerged as the cynosure of attraction among the
enthusing investigator’s thanks to the incredible efficiency in performance in the domain of the
biometric detection systems [1]. An innovative multimodal biometric system which is well-
geared to assuage the constraints of the unimodal biometric systems by integrating the data
from the related biometric sources is efficiently envisaged [7]. The novel systems are compe-
tent to effectively tackle the certain pressing issues faced by furnishing multiple pieces of proof
of the matching identity [2]. In relation to the single biometric verification techniques, the
multimodal methods boast of leading gains outsmarting those of the former. In fact, the
Multimedia Tools and Applications

multimodal biometric is well-equipped with the skills of combining several single biometric
verification techniques, amassing the merits of the entire categories of the single biometrics so
as to enhance the efficiency in accomplishment of the system and attain a further vigorous
technique [27]. Of late, the multimodal biometrics has emerged as a vital investigation
phenomenon in fine-tuning the biometric precision. For instance, multiple biometric traits
may be captured by diverse sensors and a feast of biometric processes may be performed to
yield an assortment of data and further divergent matching techniques can be designed to
achieve autonomous match scores [6]. These days, the multimodal biometrics technology uses
to multiple categories of biometrics data has well-established as an endearing approach
extensively employed for the purpose of person authentication and verification [5]. The main
contribution of the work is listed below;

& For human authentication, fingerprint and ear images are used.
& The shape feature of fingerprint and ear image is extracted with the help of the modified
region growing algorithm (MRG).
& The texture feature of both ear and fingerprint is calculated using local Gabor Xor pattern
(LGXP).
& The ear and fingerprint are recognized with the help of optimal neural network. Here, to
improve the performance of neural network, the weights are optimally selected with the
help of firefly algorithm.

The rest of the paper is organized as follows; the literature survey of proposed methodology is
explained in Section 2 and motivation of the study is given in Section 3. The proposed
methodology is described in Section 4 and result and discussion of the proposed system is
explained in Section 5. Finally, the conclusion is presented in Section 6.

2 Review of recent researches

The biometric techniques are deals with some challenging safety issues; hence they provided
some detrimental error rates at the time. The multimodal biometric systems are affected greater
part of the consistent deficiencies. Multiple biometric modalities are integrated into a single
detection technique through the multimodal biometrics and quintessence. The detection
precision was scaled up by the single detection technique. In this section, we are discussed a
few of the modern techniques,

& Desong Wang, Jianping et al. [23] have presented a multilayered backpropagation algo-
rithm. It was based on the artificial neural network (ANN) that was used for the determi-
nation of biometric gait classifications. In their article gait biometric was classified into
four types such as Normal, crouch-2, crouch-3 and crouch-4. The multilayered
backpropagation algorithm produces a gait identification with better activity recognition
accuracy (ARA). When maintaining a high ARA their method was varying for different
environments e.g. outdoor, indoor. The basic concern of their research was the determi-
nation of disorder and abnormality in gait patterns. They have also presented an early
detection methodology for abnormal gait. That was used to identify the potential diseases
from human walking. Their method also utilized for determination of some biometrics
such as gender, age, race, and authentication.
 Multimedia Tools and Applications

& Md. Maruf Monwar et al. [17] have presented a Deep Neural Network (DNN) for
determination of human push recovery data classifications. Those data’s were achieved
by applying empirical mode decomposition on various leg joint angles from the intrinsic
mode functions. During the analysis of the human push recovery mechanism; there were
four types of pushes applied such as small, medium, moderately high and high. Their DNN
was used to perform various types of pushes. Based on their technique, they have obtained
88.4% accuracy. The stable implication of the result was shown from the analysis of
variances. The consistent approaches are hip, knee and ankle were used for the various
types of pushes such as small, medium, moderately high, high. Those pushes were
determined based on the push recovery.
& Mingxing He et al. [8] have presented a ‘robotic gait recognition system for determination
of features and the principles of features. They classify gait data by applying various
machine learning techniques such as KNN, ANN, SVM, DNN, and classifier fusion. In
their article, features were selected by applying the incremental feature selection strategy.
The incremental feature selection (IFS) and analysis of variance (ANOVA) were the new
feature selected methods. From the feature selection method, generalization accuracy was
provided. Various old features were selected by using ANOVA with IFA. Then the selected
features were fed into the input of various machine learning techniques such as KNN,
ANN, SVM, and DNN. Those classifier models were used to create a classifier fusion
model. Their presented method was checked out by applying 5-fold validation method.
From the above evolution, they were providing 92.23% better result compared with
individual classifier methods. The numerical implication of their result was obtained by
applying a one-way analysis variance test, Friedman’s test and Kruskal-Wallis test. Their
method also used for Biometric and help to identification of Parkinson disease.
& Hong Huang et al. [9] have presented modeling joint trajectories of biped locomotion for
designing of a bipedal robot. That was stunning block before and after impacts. The main
aim of their article was to develop a bipedal walk and hybrid automata. They also construct
a hybrid automata dynamic walk model for individual subjects. An OpenSim bipedal
gait2354 and normal walk model as to with trajectory generated model. In their article,
they have proposed a new bipedal locomotion modeling technique by applying hybrid
automata. The hip, knee and ankle trajectories were combined from that model. The
stability of the margin was logically determined. Furthermore, that trajectory has delivered
to a real humanoid robot HOAP2. That has able to achieve stable walking.
& Li Yuan and Zhi chun Mu et al. [29] have presented a fuzzy vault based new chaff point
method. Their method was used for both the face and ear. The purpose of that method was
to enhance the security of biometric data. For that, a new chaff point feature vectors as to
with the grouped feature vectors. Those grouped vectors were created by interfacing with
extracted shape and texture feature vectors. The chaff matrixes of x and y coordination
were created by using distinct part of extracted feature vector points. From the feature
points, they were selecting a new chaff point based on the best vector points area. They
also utilize a particle swarm optimization (PSO) algorithm for the generation of optimal
locations. The particles were obtained from the extracted features location in PSO. From
that best location, based on the fitness value chaff feature point was selected. Their method
was presented on the Yale face and ear databases and performances were calculated by
applying some measures such as Jaccard coefficient (JC), Genuine Acceptance Rate
(GAR), False Matching Rate (FMR), Dice Coefficient (DC) and False Non-Matching
Rate (FNMR). The results of their evaluations provide 90% for person recognition.
Multimedia Tools and Applications

& Tak-Shing Chan et al. [3] have presented a fusion framework for feature levels. That
method was used to save the users multiple templates with a single secure sketch. The
main aim of their article as (a). The practical executions of their presented method must use
two well-known biometric cryptosystems such as fuzzy vault and fuzzy commitment. (b).
Accuracy and security tradeoff matching must be based on two various databases such as
one real and one virtual multimodal database. Each database consists of three famous
biometric modalities such as fingerprint, iris (Eyes), and face. The multi-biometric
cryptosystems presented, that provide greater security and corresponding performances.

Pflug and C. Busch et al. [18] have presented a cancelable multi-biometric system for
fingerprint and finger-vein. That was used for revocability and template security. Their
presented multi-biometric system interfaces with the feature sets of minutiae-based
fingerprint and image-based finger-vein. Furthermore, they also generated three fusion
options with fusion strategy for feature levels. Those various fusion options were
analyzed and calculated by using similar performances and security forces. Their pre-
sented method EP-DFT highly improves system security correlated with the original P-
DFT.

3 Motivation of the research

In latest years, Biometric-based methods have emerged as the most promising option for
identifying individuals. Fundamentally, we recognize the people applying passwords, PINs,
smart cards, tokens, keys etc. however these are sometimes misplaced, forgotten, purloined or
duplicate and hard to remember. But an individual’s biological traits can never be misplaced,
forgotten, stolen or forged. Biometric-based technologies comprise identification based on
physiological features (such as face, fingerprints, finger geometry, hand geometry, hand veins,
palm, iris, retina, ear and voice) and behavioral characteristics (such as gait, signature and
keystroke dynamics) [10]. More newly, finger print recognition is commonly applied in the
biometric technology [12–14]. By examining the trivia of a human being Fingerprint individ-
uality can be described. However this biometric containing the limitation is public perception
and privacy concern of criminal implication. In addition, the face is a famous biometric to
recognizing the human. It is automatic techniques to record the geometry of distinguish
characteristics of the face. Among different vendors, different techniques of facial recognition
all spotlight on measures of main features. But in case of face recognition, there can be
difficulties with e.g. sensitive to changing lightning, dissimilar head positions of the person
and expression. Along with, in addition the facial images vary with time. Similarly, ear is most
used biometric which is also used for recognition. But this also have some difficulties. To
overcome the problem, in this paper we utilized multi-biometric based authentication process.
This method no one can steal the sensitive information. so, in this paper we utilized combined
ear and face based biometric for authentication.

4 Proposed human authentication methodology

The main objective of this paper is to authenticate the human based on ear and fingerprint
biometrics. In the earlier investigation, the multimodal biometrics of Fingerprint and ear based
 Multimedia Tools and Applications

on fuzzy vault is carried out, with the intention of offering an integrated biometric system. In
the related investigation, the Fuzzy vault was employed for the safety functions, but the
deficiency was that the biometric data would be visible, thereby offering itself for the effortless
hacking. Further, the chaff points which were created arbitrarily from the biometric features
could be easily located by the assailants. With an eye on scaling up the safety aspects by means
of concealing the secret key within the biometric data, an innovative chaff points based Fuzzy
Vault is kick-started in the current investigation. The novel technique flows through the
following stages. (1) Pre-Processing. (2) Feature Extraction (3) Generation of grouped feature
vectors (4) Recognition. Initially, the input fingerprint and ear images are pre-processed with
the help of the median filter for eliminating the noises and blur from images. Subsequently, by
means of the Modified Region Growing (MRG) approach and Local Gabor XOR Pattern
(LGXP) technique, the shape and texture features of the ear and fingerprint are effectively
extracted. For the reorganization, optimal neural network is utilized. The overall diagram of
proposed methodology is given in Fig. 1.

a. Preprocessing

With the intention of achieving augmented outcomes for the authentication, the image Pre-
processing is effectively carried out. The fingerprint and ear images are physically cropped in
this task. In the ear images, the areas which do not form part of the ear is colored like black and
in the fingerprint images, the areas which are not the segments of the Fingerprint are colored as

Biometric
database

Preprocessing using median filter

Feature extraction

MRG
LGXP

Feature vector

Training Testing

Optimal weight
Firefly Using optimal
algorithm Image features
neural network
Random weight
Weight
NN

Recognized image Non

Recognized
image

Fig. 1 Block diagram for the proposed Firefly Biometric image classification
Multimedia Tools and Applications

black. The colored images, in turn, are modified to the grayscale images. The median filter is
effectively used for the elimination of noises and blur from the images.

i. Median Filter

The median filter represents a nonlinear digital filtering approach, habitually employed to get
rid of the noise. The corresponding noise cutback constitutes a characteristic pre-processing
measure to fine-tune the outcomes of the subsequent processing. In accordance with the
brightness (intensity), the neighboring pixels are graded and the median value turns into a
novel value for the central pixel. The median filters to execute an excellent task of doing away
with certain categories of noise, especially the “shot” or impulse noise wherein certain
individual pixels are endowed with fabulous values. Based on the intensity, the pixel values
in the neighborhood window are ranked in the median filtering function, and the middle value
(the median) emerges as the output value for the 0.



r
r 
Y ðt Þ ¼ median x t þ ; xðt−T 1 þ 1Þ; …xðt Þ; :…; x t þ ð1Þ
2 2
Here t is characterized as the size of the window in the median filter. Subsequently, the one-
dimensional median filter is furnished as illustrated above. In the image processing, n two-
dimensional filters are effectively employed. Generally, the images are symbolized in discrete
form as two-dimensional arrays of image elements, or pixels - i.e. sets of non-negative values
Bij ordered by two indexes i = 1,…, Ny (rows) and j = 1,…,Ny (column). Where the elements
Bij are scalar values. Further, there are a number of techniques for processing the color images,
where each pixel is denoted by various values such as “red”, “green”, “blue” values deciding
the color of the pixel.

1. The Median filtering involves a nonlinear technique employed to eliminate the noise from
the images.
2. It is extensively employed as it is very efficient in eliminating the noise concurrently
safeguarding the edges.
3. It is especially competent to steer clear of the ‘salt and pepper’ category of noise.
4. The median filter functions by traversing through the image pixel by pixel, substituting
each value with the median value of adjoining pixels.
5. The model of the neighbors is known as the “window”, which slides, pixel by pixel
throughout the whole images.
6. The median is evaluated by initially sorting the entire pixel values from the window in the
numerical order, and thereafter substituting the pixel being taken into account with the
middle (median) pixel value.

b. Feature Extraction

In this stage, the shapes and texture characteristics of the Fingerprint and ear images are
effectively extracted. The grayscale images are furnished as input to the feature extraction
procedure. The shape of the fingerprint and ear images is extracted with the help of the
modified region growing algorithm. The consistencies of the images are extracted by the
LGXP technique. Here, various features such as the shape feature, texture feature, and ear and
fingerprint image are extracted.
 Multimedia Tools and Applications

i. Modified region growing algorithm

In the modified region growing algorithm, the threshold of the image is ignored and in its
place, the threshold of the direction image is considered for the purpose of the region rising
procedure. The vital advantage in the performance of the modified region growing is related to
the fact that the shape of the image is segmented proficiently and it is possible to gather
additional data when compared with the region growing algorithm. For eliminating the
apparent segment of the ear known as the pinna and offering further data on the ear shape,
the modified region growing method emerges as highly proficient in relation to the region
growing algorithm. Now, the total number of pixels in the image is evaluated, which is
identical to the size of the image. The Gray level of the selected region in the image is
furnished by the proportion of the total of gray level for all pixels in the region to the total
number of pixels in the region:
T otal of gray level f or all pixel in the region
GrayLevel ¼ ð2Þ
T otal no of pixel in the region

ii. Local Gabor Xor Pattern

The Fingerprint and ear images were identified by means of the Gabor filter which was kick-
started initially by Dennis Gabor for 1-D signals and Daugman enlarged its scope to cover 2-D.
The Gabor uniqueness was deemed as highly effective and appropriate for the purpose of the
Fingerprint recognition. For the feature extraction, the Gabor filters and the band pass were
deployed. A bank of filters was effectively used with divergent orientations to extract the
frequency data and thereby the characteristics at different orientations because the entire facial
attributes were not available at the same orientation. At every orientation, Scaling was carried
out with the intention of attaining the maximum frequency data at every orientation. In other
words, both the orientation and scaling lent a helping hand in the extraction of the maximum
frequency data. In the LGXP, phases were at first quantized into divergent variety, thereafter
the LXP operator was deployed to the quantized phases of the central pixel and every one of its
neighbors. At last, the consequential binary labels were concatenated jointly as the local model
of the central pixel. The corresponding extracted features were integrated and classified with
the assistance of the firefly technique.

4.1 Authentication using optimal neural network

After the feature extraction process, the extracted features are given to ONN to recognize the
corresponding image is available in the database or not. Here, the data based divided into two
phase namely, training and testing. For training process, the weight values are optimally
selected. Basically, the neural network is consisting of three layers, namely, input layer, hidden
layer and output layer. The input layer and extracted features are ideal. To improve the
performance of the ANN, the weight values are optimally selected with the help of firefly
algorithm. The step by step process of optimal weight selection process is explained below;

Step 1: Solution representation

Multimedia Tools and Applications

Solution initialization is an important process for all optimization algorithms. In NN, initially,
the weight values are randomly assigned. However, for optimization initially, the solution is
randomly assigned. Here, the solution is considered as firefly and weight values are considered
as light.

Step 2: Fitness evaluation

After the solution initialization, fitness of each solution is calculated. The Fitness function is
defined in accordance with the target of the current investigation. Here, an optimization
formula is obtained in Eq. (3), dependent on the minimization of the objective function as
detailed below.
1 n ^ 2
MSE ¼ ∑ Y i −Y i ð3Þ
n i¼1

Where,
n
^
MSE represents the 1n ∑ square of the error ðY i −Y i Þ2. This is an easily evaluated quality for
i¼1
a specific sample.

Step 3: Firefly updation

The movement of the firefly (FF) p, when attracted to a different and more attractive (brighter)
firefly q, is estimated by means of Eq. 4 given below.
 
0   1
up ¼ up þ γ ðrÞ* up −uq þ φ rand− ð4Þ
2

The second term in Eq. (4) arises on account of attraction, the third term introduces the
randomization with 'φ' being the randomization parameter and “rand” represents a random
number produced evenly disseminated in the range of 0 to 1.

Attractiveness; γ ðrÞ ¼ γ 0 e−θ r ;

m
m ≥1 ð5Þ
Where, r represents the distance between two fireflies, γ0 denotes the initial attractiveness of
firefly and θ reveals the absorption coefficient.
  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2
Distance; rpq ¼ up −uq  ¼ ∑dk¼1 up;s −uq;s ð6Þ

Where, up, s represents the sth component of the spatial coordinate of the pth firefly and d
denotes the total number of dimensions. Also q ∈ {1, 2, …, Fn} characterizes the arbitrarily
selected index. Although q is evaluated arbitrarily, it has to be different from p. Here Fn
corresponds to the number of fireflies.

Step 4: Termination

The iteration is stops, when the firefly gets optimal weight. The selected optimal weight is
given to the neural network.
 Multimedia Tools and Applications

4.2 Testing process

After the training process, the testing process is carried out with the help of neural network.
Here, initially, the feature vector of both ear and fingerprint is given to the input of the neural
network. The trained weight value is directly assigned to the NN. Based on the neural network
score value, the classifier classify the image is recognized or not.

5 Results and discussions

The innovative firefly technique for the proficient classification of the biometric cryptosystem
for the Human authentication employing the Fingerprint and Ear in the same individual is
performed in the MATLAB platform with the Fingerprint and Ear images collected from the
dataset (Fig. 2).

5.1 Performance evaluation

With the assistance of the efficiency metrics like the Sensitivity, Specificity, and Accuracy, the
efficiency in accomplishment of the system is effectively evaluated. The fundamental count
values including the False Positive Rate (FPR), False Negative Rate (FNR), Sensitivity, and
the Specificity, is effectively employed by the relative metrics. The categorization of finger-
print and ear images together with the classification by means of efficiency is evaluated by the
novel technique, which is comprehensively dealt with in the ensuing sections.

& False Positive Rate (FPR)

It represents the percentage of cases wherein an image is categorized as belonging to the

identical person images, though, in actual fact, it does not belong. It is evaluated by means of
Eq. 7 shown below.

Fig. 2 Experimental sample images

Multimedia Tools and Applications

FP
FPR ¼ ð7Þ
FP þ TN

& False Negative Rate (FNR)

It symbolizes the percentage of cases wherein an image is categorized as not belonging to the
identical person images, though, in reality, it belongs to the related images. It is estimated in
terms of the ensuing Eq. 8.
FN
FNR ¼ ð8Þ
FN þ TP

& Sensitivity

It is defined as the percentage of the number of the actual positives which are appropriately
categorized to the total number of actual positive and false negatives. It is a measure intended
for evaluating the skills in recognizing the positive outcomes. It is determined as illustrated in
Eq. 9 shown hereunder.
Number of true positives
Sensitivity ¼  100 ð9Þ
Number of true positives þ Number of false negatives

& Specificity

It is defined as the percentage of the number of the actual negatives which are appropriately
categorized to the total number of the actual negatives and false positives. It is a metric used for
the purpose of appraising the skills in recognizing the negative outcomes. It is evaluated by
means of the following Eq. 10.
Number of true negatives
Specificity ¼  100 ð10Þ
Number of true negatives þ Number of false positives

& Accuracy

Accuracy is defined as the percentage of the sum total number of actual positives and actual
negatives to the total number of images comprising both types of positives and negatives. It is
employed to effectively appraise the metric of accuracy from the metrics of sensitivity and
specificity as illustrated in Eq. 11 shown below.
TP þ TN
Accuracy ¼  100 ð11Þ
TP þ TN þ FP þ FN
 Multimedia Tools and Applications

Equations (6–10) shown above are the appropriate equations for the purpose of evaluating the
efficiency in the classification of the images.

5.2 Performance analysis

Invariably on all occasions, images are classified for the fingerprint and ear images.
The effectiveness of the related classification of the entire images appraised by means
of the performance metrics such as the False Positive Rate, False Negative Rate,
Sensitivity, Specificity and Accuracy as elegantly exhibited in Eqs. (6–10).
The classification of the biometrics images has brought in brilliant accuracy
upshots. The following Fig. 3 beautifully depicts the relative graph in respect to the
values in Table 1.
Table 1 shown above together with the relative graph in Fig. 3 effectively exhibit
the classification efficiency values. It is observed that the FPR values are the least for
our innovative technique, which goes a long way in significantly stepping up the
Classification accuracy. The maximum specificity for the dataset is achieved in
iteration 25 with a value of 0.9661, and the related figures for the iterations 50, 75,
and 100 are 1, 0.9553 and 1 correspondingly. As the corresponding values achieved
by the innovative firefly algorithm are appreciably elevated, they act as a boost for
the surge in accuracy. However, the sensitivity values for the five images are the least
with values of 0.7, 0, 0.9625, and 0.0125, for iterations 25, 50, 75, and 100 in that
order, though the least values do not in any way affect the excellence of the
classification. With the result, our ground-breaking technique comes out in flying
colors realizing superb accuracy values of 0.9628, 0.9810, 0.9875%, 0.9554 and
0.9876% for the iterations 26, 50, 75, and 100, correspondingly. Thus, it is heartening
that the epoch-making approach has accomplished its innate skills in ushering in an
accuracy of 0.98% for the classification.

Fig. 3 Evaluation metrics for Accuracy, sensitivity, Specificity

Multimedia Tools and Applications

Table 1 Biometric image classification in various images

FPR FNR GAR Specificity (in %) Sensitivity (in %) Accuracy (in %)

0.0338 0.0039 0.9962 0.9661 0.7 0.9628

0 0.0125 0.9875 1 0 0.9875
0.0446 0.0004 0.9995 0.9553 0.9625 0.9554
0 0.0123 0.9876 1 0.0125 0.9876

5.2.1 Comparative analysis of our proposed work with the existing works

To prove the effectiveness of the proposed methodology, our proposed method is compared with
three existing method namely, Semwal et al. [24], Semwal et al. [25] and Semwal et al. [26]. In [24],
ANN based biometric gait identification is performed. Here, for feature extraction process kernel
principle component analysis is used (KPCA). In [25], an optimized feature selection technique
based on incremental feature analysis for bio-metric gait data classification is developed. Similarly,
in [26], Human Gait State Prediction Using Cellular Automata and Classification Using ELM are
developed. Table 2 shows comparative analysis of proposed against existing.
In Table 2, we compared performance of proposed method with different method. When
analyzing Table 2, our proposed method achieves the maximum accuracy of 97.33% which is
90.23% for using Semwal et al. [24], 93.56% for using Semwal et al. [25] and 63% for using
Semwal et al. [26]. Similarly, we obtain the average sensitivity of 84.6% which is 65.32% for
using Semwal et al. [24], 72.8% for using Semwal et al. [25] and 70.12% for using Semwal
et al. [26]. Moreover, we obtain the maximum specificity also. This because of, in this paper
we utilized hybrid modality and also we utilize optimal neural network. From the result, we
clearly understand our proposed methodology is better than other method.

6 Conclusion

An innovative optimal neural network based biometric image classification method with three
diverse phases such as the Preprocessing, Feature Extraction and classification is elegantly launched
in this document. The novel technique commences with the task of preprocessing involving the
median filters. It is followed by the feature extraction stage in which various features are effectively
extorted from the biometric images. The extracted features include the shape and texture feature like
the fingerprint and ear feature. The ONN technique admirably classifies the images. The efficiency
metrics such as the False Positive Rate, False Negative Rate, sensitivity, specificity and accuracy are
successfully estimated for the new-fangled technique, which comes out in flying colors in classifying
the images with superlative efficacy, thereby accomplishing sterling efficiency in the task of
classification of images and turning out cheering outcomes of amazing accuracy.

Table 2 Comparative analysis

Methods Avg. Accuracy (%) Avg. Sensitivity (%) Avg. Specificity (%)

Proposed 97.33 84.6 98.03

Semwal et al. [24] 90.23 65.32 75.67
Semwal et al. [25] 93.56 72.8 80.23
Semwal et al. [26] 63 70.12 54
 Multimedia Tools and Applications

References

1. Ahmad M1, Woo WL, Dlay SS (2010) Multimodal biometric fusion at feature level: face and palm print. In:
Proc. of the 7th international symposium on communication systems networks and digital signal processing
(CSNDSP), pp 801–805
2. Basha AJ, Palanisamy V, Purusothaman T (2010) Fast multimodal biometric approach using dynamic
fingerprint authentication and enhanced Iris features. In: Proc. of the IEEE international conference on
computational intelligence and computing research (ICCIC), 2010, pp 1–6
3. Chan T-S, Kumar A (2012) Reliable ear identification using 2-D quadrature filters. Pattern Recogn Lett
33(14):1870–1881
4. Chin YJ, Ong TS, Teoh ABJ, Goh MKO (2011) Multimodal biometrics based bit extraction method for
template security. In: Proc. 6th IEEE conference on industrial electronics and applications (ICIEA), pp
1971–1976
5. Choi H, Shin M (2009) Learning radial basis function model with matching score quality for person
authentication in multimodal biometrics. In: Proc. of first Asian IEEE conference on intelligent information
and database systems, pp 346–350
6. Dahel SK, Xiao Q (2003) Accuracy performance analysis of multimodal biometrics. In: Proc. of the IEEE
workshop on information assurance, pp 170–173
7. Hanmandlu M, Grover J, Gureja A, Gupta HM (2011) Score level fusion of multimodal biometrics using
triangular norms. Pattern Recogn Lett 32(14):1843–1850
8. He M, Horng S-J, Fan P, Run R-S, Chen R-J, Lai J-L, Khan MK, Sentosa KO (2010) Performance
evaluation of score level fusion in multimodal biometric systems. Pattern Recogn 43(5):1789–1800
9. Huang H, Liu J, Feng H, He T (2011) Ear recognition based on uncorrelated local fisher discriminant
analysis. Neuro Computing 74(17):3103–3113
10. Huang Z, Liu Y, Li C, Yang M, Chen L (2013) A robust face and ear based multimodal biometric system
using sparse representation. Pattern Recogn 46(8):2156–2168
11. Ichino M, Sakano H, Komatsu N (2006) Multimodal biometrics of lip movements and voice using kernel
fisher discriminant analysis. In: Proc. of the 9th international conference on control, automation, robotics
and vision (ICARCV '06), pp 1–6
12. Islam SMS, Davies R, Bennamoun M, Owens RA, Mian AS (2013) Multibiometric human recognition
using 3D ear and face features. Pattern Recogn 46(3):613–627
13. Khan MK, Zhang J (2008) Multimodal face and fingerprint biometrics authentication on space-limited
tokens. Neuro Computing 17(12):3026–3031
14. Kumar A, Chan T-ST (2013) Robust ear identification using sparse representation of local texture descrip-
tors. Pattern Recogn 46(1):73–85
15. Madhavi, Jain AK, Flynn P, Ross AA (2007) Handbook of biometrics. Springer
16. Maple C, Schetinin V (2006) Using a Bayesian averaging model for estimating the reliability of decisions in
multimodal biometrics. In: Proc. of the first IEEE international conference on availability, reliability and
security, pp 929–935
17. Monwar MM, Gavrilova ML (2009) Multimodal biometric system using rank-level fusion approach. IEEE
Transactions on Systems, Man and Cybernetics 39(4):867–878
18. Pflug A, Busch C (2012) Ear biometrics: a survey of detection, feature extraction and recognition methods.
IET biometrics 1(2):114–129
19. Raghavendra R, Imran M, Rao A, Hemantha Kumar G (2010) Multimodal biometrics: analysis of Handvein
& Palmprint Combination Used for person verification. In: Proc. of third IEEE international conference on
emerging trends in engineering and technology, pp 526–530
20. Rahman MM, Islam MR, Bhuiyan NI, Ahmed B, Islam MA (2007) Person identification using ear
biometrics. International Journal of The Computer, the Internet and Management 15(2):1–8
21. Ross A, Abaza A (2011) Human ear recognition. IEEE Computer Magazine 44(11):718–737
22. Ross A, Jain AK (2004) Multimodal biometrics: an overview. In: Proc. of 12th European signal processing
conference (EUSIPCO), vol 14, no 1, pp 1221–1224
23. Seal DB, Nasipuri M, Basu DK (2011) Minutiae based thermal face recognition using blood perfusion data.
In: 2011 International Conference on Image Information Processing, Shimla, pp 1–4
24. Semwal VB, Raj M, Nandi GC (2014) Multilayer perceptron based biometric GAIT identification. Robot
Auton Syst
25. Semwal VB, Singha J, Sharma PK, Chauhan A, Behera B (2017) An optimized feature selection technique
based on incremental feature analysis for bio-metric gait data classification. Multimed Tools Appl 76(22):
24457–24475
26. Semwal VB, Gaud N, Nandi GC (2019) Human gait state prediction using cellular automata and classifi-
cation using ELM. In: Machine intelligence and signal analysis. Springer, Singapore, pp 135–145
Multimedia Tools and Applications

27. Yang JC (2010) Biometrics verification techniques combing with digital signature for multimodal biomet-
rics payment system. In: Proc. of IEEE international conference on management of e-commerce and e-
government, pp 405–410
28. Yaoa Y-F, Jing X-Y, Wong H-S (2007) Face and palm print feature level fusion for single sample biometrics
recognition. Neuro Computing 9(8)
29. Yuan L, Mu Z c (2012) Ear recognition based on local information fusion. Pattern Recogn Lett 33(2):182–190

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Padira.S.V.V.N.Chanukya (padira.sathya.veera.venkata.naga.chanukya) obtained his Bachelor’s degree in Elec-

tronics and communication engineering from University of JNTUK. Then he obtained his M.E in Embedded
systems and PhD in image processing, biometrics from University of MAHER University. Currently, he is a
lecturer at the Faculty of Nalanda Institute of Engg&technology (affiliated JNTUK). His specializations include
biometrics and pattern recognisation. His current research interests are image processing, signal processing.

T.K Thivakaran (Thorapadi Kannappan Thivakaran) has completed B.E in Electrical Engineering from Bangalore
University and M.E in Computer science and Engineering from Annamalai University. He has completed his PhD in
Computer science and Engineering from Manonmanium Sundaranar University. He has completed is MBA in
marketing from University of Madras. He is working as a professor in the Department of Computer Science and
Engineering, Sri Venkateswara College of Engineering, Sriperumbhudhur Chennai. He has published papers in
International and National Journals cited in Scopus. His research area of interest are Image Processing, Cryptography,
Data mining, Wirless adhoc and sensor networks, Management Information systems.