International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 4 Issue: 7 76 - 79
______________________________________________________________________________________
Texture Features from Handwritten Images for Writer Identification

Rashmi Mundas Deepa Bendegeri

Mtech In Information Technology, ISE Department Assistant Professor, ISE Deparment
SDM College Of Engg & Tech., Dharwad SDM College Of Engg & Tech., Dharwad
KARNATAKA, INDIA KARNATAKA, INDIA
rashmism232@[Link] deepabendigeri@[Link]

Dr. Jagadeesh Pujari

HOD, ISE Deparment
SDM College Of Engg & Tech., Dharwad
KARNATAKA, INDIA
jaggudp@[Link]

Abstract Identification of the writer is having wide scope in emerging technology due to its usage in various types of applications, especially
in forensic science and biometric science. Our aim in this project is to identify author or writer from script which is handwritten and obtained as
scanned images. Features of textures will be elicitated from wavelet decomposed images based on co-occurrence histograms. These will get
(capture) the information about the relations among sub-bands of less frequency and that in sub-bands of higher frequency at the particular level
of the transformed image. If the co-relation between the sub-bands has resolution of same then that indicates a stronger relation. Then
relationship strength will indicate as information was essential considered to differentiating the textures. The proposed methodology will be
executed with English handwritten images by considering 5, 10 penmanship or writers. Ability of features from texture in identifying writers is
indicated though the outcome achieved in experimentation.

Keywords-Wavelet; Texture features;Writer identification; document image; scanned images; co occurrence histograms
__________________________________________________*****_________________________________________________

that they utilize the information of the substance of the

I. INTRODUCTION information to isolated style from substance. This will build the
Writer identification is a framework designed for precision of text-dependent frameworks. The significant issue
recognizing the writer of a manually written record. An of text-dependent frameworks is that they are non-relevant to
arrangement of reports from known authors must be known situations when the content is not accessible, for example, in
ahead of time to select another archive to one of this writer. To criminal equity frameworks includes examination between
begin with, components (features) are registered on the writer content archives with various substance. Also, message
of a reference report and after that these feature parts will be subordinate frameworks are much inclined to falsification when
appeared differently in relation to the ones which exist in the same information is displayed during testing. These sorts of
database set. An author having most noteworthy closeness to frameworks can be actualized in the agreeable environment,
the existing is related to archive. Writer identification where significant concern is exactness and writer might be
distinguishes the handwriting methodology by taking into asked to author particular substance to demonstrate their
account of unknown handwriting image which continues by personality. These strategies are fundamentally the same as
coordinating obscure handwritten image against a database of signature check procedures which includes the correlation
composed examples with known writer. In this manner, writer between individual characters or words which have known
identification is imperative with numerous applications, for semantic substance. Subsequently these techniques require
example, report examination, security, and monetary action, earlier restriction and division of the right data, which is
measurable and utilized as access control. The difficulties for generally performed by human client collaboration.
identification of the writer and retrieving the author The text-independent writer identification framework
incorporates from various pen features, that differs in the displays the style data, free of the content, which is utilized to
writer handwriting style of their composition, if certainty recognize the writer in view of any given text content. This
which the author (writer) has composed content in the rush or requires the insights of features which are computed from
not, furthermore that single word is uncommon composed the large extensive quantity of data to avoid anomalies due to
very same way twice. specific content. Writer identification and verification
Writer identification technique falls in classifications: Text technique utilizes measurable elements which are extracted
dependent and text independent. Text-dependent approaches from entire image containing a text segment (block). Base
require handwritten tests taking into account on a particular measure of manual written (e.g, few content lines in section) is
content, or expect handwriting recognizer accessible for extremely important to determine stable features components
checking realness of writer. Writer identification utilizing which will be hard to the text content of the examples.
signature is most prominent occurrence of these sorts of Consequently our technique will fall in this later class.
methodologies. Text-dependent approaches have advantage
76
IJRITCC | July 2016, Available @ [Link]
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 4 Issue: 7 76 - 79
______________________________________________________________________________________
Writer identification includes input of two types: on- carry out down sampling. This transform decomposes an image
line and off-line. Online technique includes catching of pen with the overall scale factor of four, providing at each level one
development of author, where the style is compelled. Online low resolution subimage and three wavelet coefficient
writer identification framework uses temporal succession subimages
(sequence) code, which tracks pressure and velocity(speed)
varies in handwriting, and pattern(shape codes) that relay on A = |Sx * | Sy * I |2,1|2,1
direction of trajectory in writing was developed for Chinese
and English language [Bangy Li and Tieniu Tan 2009]. It H = |Tx * | Sy * I |2,1|2,1
works better for little number of characters. Online text-
independent writer recognition framework [PitakThunswarin V = |Sx * | Ty * I |2,1|2,1
and TakenobuMatssura] for the language Thai depends on
speed of pen pointer utilizes Fourier change technique. D = |Gx * | Gy * I |2,1|2,1
In offline writer identification framework, scanned image of
the author composing is utilized which delineates his Here I is the input image. Sx, Sy and Tx, Ty represent low
behaviour. Offline text independent writer identification using and high pass filters respectively, * denotes the convolution
Hidden Markov Model [Andreas Schlapbach and Horst Bunke] operator and 2 denotes downsampling operation. The subbands
works on the basis of computing the score unknown author and labeled H, V, D correspond to the detail images, while A
comparing it scores of every individual author. The score of corresponds to the approximation image as shown in the Fig. 1.
every individual author is computed by recognizer in view of
hand writing. The recognizer with the most astounding score is
assigned as unknown author. In offline writer identification
framework, the hand written text of the writer is filtered and
utilized for feature extraction. In that capacity offline writer
identifications postures more difficulties contrasted with on-
line method on account of the absence of extra features, which
are accessible to online frameworks, is absent for offline
system. Statistical based Writer identification method for non
consistently skewed handwriting images has been discussed in
(a) (b)
[H S Said, K Baker and also T N Tan]. Different strategies for
writer identification depends on Contour based features,
Hierarchical Shape Primitive elements, has been talked about in
[Mohamed NidhalAbdi and Maher Khemakhem]. Offline text-
independent writer identification is very imperative for B. Computation scheme
measurable examination, archives approval, and calligraphic Script or style types usually vary from each other by the
relics ID. way they are assembled or grouped into words, and also the
state of individual attribute, and so forth. This gives diverse
II. FEATURE EXTRACTION scripts particularly distinctive visual appearance. Texture can
A. Discrete wavelet transform be characterized in simple definitive form as same pattern
The continuous 1-D transform wavelet of (1D) signal occurring repetitively or something comprising of commonly
F(m) described as related elements. This identification of script or writer style
(WaF)(b) = F(x)*a,b (x)dx from the handwritten images consist features which are based
represents wavelet calculated according to Base on texture, extracted from handwritten images provided by
wavelet interpretation, expansion writer in English Language The feature extraction method is
x,y(M) = 1/ |a| (x-a/b) described below.
Due to certain conditions, base wavelet fulfills limitation
which includes zero mean. It is differentiated through control The computation scheme is extraction of features
of x,y to distinct cross section (x=2y, b l). Normally, forced it influenced by perception of individuals is talented for
changes to non excess, integrated and that it should include recognition among new writings simply in view of simple
multi-resolution primitive signal view. Expansion of Two-D is visualization analysis. Classification of texture is processed by
generally operates utilizing the result of One-D wavelet considering the Identification of script. Hence, this is
channels. The Haar wavelet is defined as complicated visible texture made out derived by sub-pattern.
(t) = {1 -1 0 0<=t<=1/2, <=t<1, otherwise Despite of fact that, the sub-patterns can have scarcity of a
Above equation is much easier, where it checks {n,a = ( 2- better mathematical standard, it is well entrenched that a
n/2
(2 n/2 s a))}n,aZ is orthogonal and unit vectors premise texture is considered as completely only if all the sub-patterns
for L2(R) . In this, discretization x=2a and y=s2a is utilized, are correctly defined. We utilize a multi-resolution method as
which will be p sought after all through this area. This wavelet elicitation of texture features according to DWT and by using
is, verifiably, primarily identified wavelet. minimum distance classifier, the classification of the textures is
The simplest way to compute a 2D discrete wavelet obtained. This extraction of features is illustrated as:
transform (DWT) of an image is to apply one-dimensional
transform over image rows and columns separately and then to
77
IJRITCC | July 2016, Available @ [Link]
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 4 Issue: 7 76 - 79
______________________________________________________________________________________

Handwritten image is considered as input image X D (N) =

and using Haar wavelet we apply 2D discrete wavelet
transform (DWT), which provides us approximation sub-band
Herein, N represents overall components in component space
image (I), and also detail sub-band images (H,V,D) (Fig 3.3).
f, fl (Y) expresses jth component of texture of given example
We take set of two images like (I, V), calculate the co
X, and fl (N) reproduces lth texture value for Mth class in
occurrence histograms H1, H2 in provided direction. For every
library. After this process, the written record is determined by
histogram, normalized cumulationis composed for all co-
minimum distance classifier.
occurrence histogram and then enumerate the texture features,
especially, mean, regression-line slope and mean-deviation IV. EXPERIMENTAL RESULTS AND DISCUSSION
described as above. The overall procedure will be looped in
Writer identification have effective approach in image
eight directions which yields two-histograms * three-features *
document analysis, this observation is considered for grouping
eight-directions = 48 components features from each set (I, V).
of composition. This approach is discussed in our project;
Correspondingly, for all sets such as, (I,H), (I,D), (I,
efficiency of feature elicitation method for texture is
abs(V-H-D)) by elicitation of features, we obtain 192
articulated. We already discussed about different writers, who
components to given handwritten image (I).
have different sense of writing in different state of mind, which
The detailed process will be repeated for the image which also includes different styles. Hence we consider, a text block
is complement image of I, represented as as distinct pattern. This study helps us to motivate the use of
where i is gray-value pixel for image I. From I and features texture classification for identification of writer. Our approach
extracted for combining and obtaining a feature space will not involve connected component method. Since, this is a
containing dimension of 384. These are used for training of global approach used for texture classification. We perform
features and then classification. Fig 3.3 illustrates the feature experiments by allowing the authors to write different scripts in
extraction schema in detail. The schematic diagram of the English language. These scripts are digitized each to 150 dpi.
feature extraction method is shown in the Fig. 2. Execution of our method is done on 10 different writers. From
each writer we have taken 10 images, from which, we
III. TEXTURE TRAINING AND CLASSIFICATION considered 5 images for training, other 5 images for testing.
A. Training
Training step includes extraction of features from different
handwritten image samples, which are chosen randomly that
belongs to every script using above feature extraction
methodology. These extracted features will be saved in the
feature library. Then these will be utilized for writer
identification.
B. Classification
Classification step involves comparison of features
values with that saved in feature library. The extraction of
features for image I is done using feature extraction method, as
explained above. Then using distance vector code these will be
compared with subsequent component values which will be
saved in library,
78
IJRITCC | July 2016, Available @ [Link]
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 4 Issue: 7 76 - 79
______________________________________________________________________________________

V. CONCLUSION
Here, we implemented a technique for extraction of texture
features from handwritten images for identification of writer.
Texture features will be extracted from wavelet decomposed
images by the use of the relationship between the sub-bands, on
the basis of co occurrence histograms. This methodology
results as a powerful differentiator among the authors or writers
comparing various manually written scanned images. The
executed results demonstrate the capacity from proposed
technique furthermore capability of the wide worldwide
methodology to distinguish proof of author in examination of
record/report investigation which includes wide significance
from the scientific as well as biometric discipline.
REFERENCES
[1] P.S. Hiremath, Shivashankar S, Jagdeesh D Pujari,
Mouneswara., Script Identification in a handwritten
document image using texture features, In Proc. IEEE 2 nd
International Advance Computing Conference, Patiala, pp.
110-114, 2010.
[2] Said, H.E.S, T N Tan and K D Baker., Personal
identification based on handwriting, Pattern Recognition,
2000.
[3] Tan T., Rotation Invariant Texture Features and their Use
in Automatic Script identification, IEEE Trans. Pattern
Analysis and Machine Intelligence, pp. 751-756.
[4] Zhu, Y., Tan, Wang Y., Font Recognition Based on
Global Texture Analysis, IEEE Trans, Pattern Analysis
and Machine Intelligence, vol 23, no 10,pp.1192-1200.
[5] Zois, E N and V Anastassopoulos., Morphological
waveform coding for writer identification, Pattern
Recognition 385-398, 2000.
[6] Hertel and Marti., A set of novel features for writer
identification in Audio and Video Based Biometric Person
Authentication, ed. J Kittler and M S Nixon, pp. 679-687,
2003.
[7] Schlapbach A., Bunke., A Writer identification and
verification system using HMM based recognizers, Pattern
Analysis Application (Springer).

79
IJRITCC | July 2016, Available @ [Link]
_______________________________________________________________________________________