Introduction

Previous work in the field of Offline Handwritten

Character Recognition

References

2

Literature Review

Chapter 2

17

Chapter 2

LITERATURE REVIEW

2.1 Introduction

Handwritten recognition is an area of research where many researchers have presented

their work and is still an area under research to achieve higher accuracy. In past

collecting, storing and transmitting information in form of handwritten script was the

most convenient way and is still prevailing as a convenient medium in the era of digital

technology. As technology has advanced tablet and many similar devices allows

humans to input data in form of handwriting. Use of paper to write handwritten text,

converting to an image using scanner, identifying handwritten characters from the

image is known as off-line handwritten text recognition is a challenging area due to the

fact that each writer will have different style of writing and all scripts have their own

character set and complexities to write text.

Many researchers have worked in the area of handwriting recognition, and numerous

techniques and models have been developed to recognize handwritten text both online

and offline. One can trace extensive work for English and Arabic script whereas

research for handwritten character recognition has progressed in identifying

handwritten characters for many Indian scripts like Devnagari, Tamil, Telugu,

Kannada, Hindi, Gurumukhi, Marathi, Gujarati.

This chapter presents review of research work carried out so far in the area of offline

handwritten character, numeral and text recognition. Algorithm and research work

varies from language script to script as many diversities exist such as number of

alphabets, alphabets used to write documents, writing direction etc. Researcher have

presented review of literature on the basis of “Language script” used for handwritten

i.e. English and several Indian scripts.

Chapter 2

18

2.2 Previous work in the field of Offline Handwritten Character Recognition

English

Dileep Kumar Patel et.al.[1] have proposed multi-resolution technique using Discrete

Wave Transform(DWT) and Euclidean distance metric (EDM) in an attempt to

recognize offline handwritten character. Various processing steps performed on

scanned handwritten image such as binarization, skeletonization, edge detection,

removal of slant angle, multi-resolution using wavelet transform, pattern classes and

mean vector, Minimum distance classifier for getting output. Authors have reported

result of 90% of accuracy with level of multi-resolution set to three and concluded that

further increment in the level of multi-resolution will decrease average recognition

accuracy.

Ravi sheth et.al. [2] have proposed chain code based feature extraction method based

on 8-neighbor principle. For feature extraction 8 different codes are derived based on

neighbor principle and later tested with Neural network (NN) and Support vector

machine (SVM) classifier.

Anita Pal & Dayashankar Singh [3] has performed a work to recognize handwritten

English character for which multilayer preceptor approach is used with one hidden

layer. Result obtained was 94% with Fourier descriptors and back propagation network

Experiment was carried out on skeletonized and normalized binary images. [4]

Sandeep saha et. al.[5] have proposed 40-point feature extraction method for

recognizing offline handwritten alphabetic character and multi-layer feed forward

neural network.

Brijesh verma et.al. [6] have presented novel approach to extract structural features

based on contour code and stroke direction for recognizing segmented cursive

handwriting. Properties such as ascender and descender count, start point and end point,

rate of change of slope are considered for contour code feature where as individual line

segments, stroke from outer boundary properties can be extracted based on directional

features. On cursive character dataset some of the process performed includes

Chapter 2

19

thresholding, slant estimation and correction, baseline computation, noise and underline

removal, segmentation. In this work authors have concluded that contour code

technique produces top results by comparing it with directional and transition feature

extraction techniques.

To recognize handwritten word Yamada and Nakano [7] has presented character

recognizer system in which segmented characters from the CEDAR benchmark

database were trained for classification. [8].

In a research presented by Kimura et al. [9] where to recognize segmented character a

52-output classifier was used along with neural and statistical classifiers. For case

sensitive experiments, their neural classifier produced an accuracy of 73.25%, which

again was comparable to the combined dataset result of 84.58% presented in this

research. [8]

In work presented by Singh and Hewitt [10] linear discriminant analysis approach is

proposed and authors have obtained a recognition rate of 67.3%. The result obtained

from the combined dataset (84.58%) compares well to their recognition rate.[8].

In [11] authors have proposed a method building skeletal graphs from handwritten

images to extract high level features i.e. loops, turns, ends, junctions. To obtain result

block adjacency graph is used as a base. Authors have tested proposed approach on

3000 word images from US postal addresses digitized at 212 dpi.

Cheng-Lin Liu et.al.[12] have proposed improved normalized functions and directional

feature extraction strategies. Dimension and moments based normalization functions

and eight feature vectors is compared on three distinct data sources and have concluded

that moment based function outperform dimension based functions.

Multiple feature extraction technique is proposed in [13] where six methods for

extracting features are used i.e. multi zoning, modified edge maps, structural

characteristics, projections, concavities, measurements and gradient directional

features. Approach was tested on MNIST database using Multilayer perceptron (MLP)

with resilient backpropagation.

Chapter 2

20

Structural features includes number of vertical and horizontal lines, end points, number

of intersections and junctions, presence of loops, number of loops, number and

intersections between the character and straight lines, holes position[ 14][15].

To recognize handwritten numerals Lee and Gomes [16] have extracted features such

as number of central, left and right cavities, location of each central cavity, the crossing

sequences, the number of intersections with the principal and secondary axes and the

pixel distribution.[15]

Sami M. Halawani et. al. [17] have proposed use of geometrical feature to recognize

hand printed characters. Low level features such as end-point, T-point and a global

information such as number of corners – a high stroke curvature. Authors have achieved

99% of recognition rate.

M. Karic and G. Martinovic [18] have proposed use of concavity features along with

gradient and chain code and for classification two Support vector classifiers are used

one with radial basis function kernel and polynomial kernel. Authors have tested work

on MNIST, USPS and USPS-r.

V Vijay Kumar et.al.[19] have divided handwritten digits into two groups. One in which

digit contains blobs with/without stems and other group of digits having stem only.

Morphological region filling approach is proposed to identify blobs. Connected

component is used to identify blobs and stems. By proposed approach average

recognition success rate is above 90% is reported.

Devnagari

For recognition of isolated handwritten Devnagari numerals Principal component

analysis (PCA) along with edge direction histogram and spines are used.[20][21].

A comparative study of Devnagari handwritten character recognition using twelve

different classifiers and four sets of feature was presented by Pal [22][21].

Chapter 2

21

B.V. Dhandra et.al. [23] have proposed neural network approach along with four

feature selection methods i.e. maximum profile distances, fill hole density, water

reservoir and directional density features for recognizing Devnagari handwritten

numerals and have achieved 98.4% recognition rate.

Mahesh Jangid et.al.[24] have applied various preprocessing steps such as intensity

adjustment, binarization, removing connected components, removing salt and pepper

noise, normalizing image to 90 x 90 on handwritten Devnagari numerals. Recursive

subdivision of character image is carried out for extracting features and for

classification Support vector machine (SVM) classifier is used and 98.98% recognition

rate is achieved. Further in [25] authors have implemented density and background

directional distribution features for various zones where every zone is of size 8 x 8

pixels and each zone contains 9 features consisting of one density feature and 8

background directional distribution features. For classification Support vector machine

(SVM) is used with RBF kernel.

To recognize non compound devnagari handwritten characters Sandhya arora et.al.[26]

have proposed approach in which two approaches are combined i.e. Multilayer

Perceptron(MLP) and Minimum edit distance. Shadow features and chain code

histogram features are used with MLP.

N. Sharma and U. Pal [27,28] proposed a directional chain code features based

quadratic classifier and obtained 80.36% accuracy for handwritten Devnagari

characters and 98.86% accuracy for handwritten Devnagari numerals.[26]

For skew correction [29] have used Wigner-Ville distribution which is a joint function

of time and frequency of the projections taken at various angles for skew angle detection

task of handwritten devnagari script with a assumption that shirorekha should be

present in document.

R J Ramteke et al [30] have presented use of invariant moments and the divisions of

image to recognize images.

Chapter 2

22

For Devnagari handwritten character recognition Kumar et.al. have proposed Zernike

moment feature based approach and artificial neural network is used for classification

of handwritten devnagari character recognition.[31]

Shailedra Kumar Shrivastava and Sanjay S.Gharde [32] have proposed affine invariant

moment and Support vector machine for recognition and have achieved average

recognition rate of 99.48%. [4]

Prachi Mukherji and Priti Rege [33][34] used shape features and fuzzy logic to

recognize offline Devnagari character recognition. They segmented the thinned

character into strokes using structural features like endpoint, cross-point, junction

points, and thinning. They classified the segmented shapes or strokes as left curve, right

curve, horizontal stroke, vertical stroke, slanted lines etc. They used tree and fuzzy

classifiers and obtained average 86.4% accuracy.

To recognize handwritten Devnagari numeral Sontakke and Patil et al. [35][34] have

proposed general fuzzy hyperline segment neural network approach. They proposed a

rotation, scale and translation invariant algorithm and reported 99.5% recognition rate.

Anilkumar N Holambe et. al.[36] have combined statistical, structural and global

transformation and moment features to have hybrid feature vector. To extract structural

features image is divided into zones and various features are extracted from zone such

as starting point, intersection point, line segment, determine line direction, type and

normalize length. Structural features based on regional properties are extracted such as

Euler number, regional area, eccentricity, orientation and for Global transformation

moments Zernike moments. Further authors in [37] have used Gradient feature as

feature extraction technique to recognize handwritten character and number of

Devnagari script. Sobel and Robert operators are used for extracting gradient features.

For classification Support vector machine (SVM) and K-nearest neighbor (KNN)

approaches are combined and have concluded that combination proves better on all

types of datasets.

Chapter 2

23

Combination of global and local features approach was presented in a work [38]] to

recognize handwritten Devnagari digits. Global features extracted includes end point,

junction point, loop, maximum profile distance, ‘U’ shape structure, ‘C’ shape structure

and to extract local features such as centroid distance feature, number of peaks in

histogram, sorted density in histogram image is divided into zones. For classification

neural network approach is used.

Gurumukhi

Sharma D. et.al. [21][39] have proposed work to recognize isolated handwritten

Gurumukhi characters where zoning is used to derive feature vector and two

classification methods were used i.e. K-nearest neighbor (KNN) and Support vector

machine (SVM) and have achieved result as 72.54%, 72.68%, 73.02% and 72.83%

respectively.

For feature extraction Singh[21][40] have extracted statistical features such as zonal

density, four directional histogram projection profile, four distance profile along with

background directional distribution (BDD) features. The highest accuracy obtained was

95.04% as 5- fold cross validation with SVM classifier used with RBF kernel among

SVM, K-NN and PNN classifiers.

In the research work by kartar Singh et. al. [41] have computed distance from outer

edges of character to bounding box of character from four sides for extracting features

from Gurumukhi numerals. Further they have created another feature vector set by

counting number of foreground pixel in specific direction in horizontal, vertical and

both diagonal direction. Image is divided into 16 zones and have considered

background directional distribution features and found best result in recognizing

Gurumukhi numerals against projection histograms and distance profile methods.

Sukhpreet singh et.al.[42] have proposed use of Gabor filters to recognize isolated

handwritten Gurumukhi characters along with Support vector machine (SVM) for

classification and tested using five-fold cross validation function and have concluded

to have 94.29% of maximum recognition rate for proposed work.

Chapter 2

24

Bangla

Ahmed shah et. al.[43] have presented their work for recognizing offline handwritten

Bangla characters. Bangla text is taken as input, segmented into lines, words and

characters, boundary detection of each character, size normalization and compared with

equivalent superimposed form.

U.Pal. et.al. [44] have presented their research work to recognize unconstrained offline

Bangla handwritten numerals. Water overflow reservoir and topological and structural

features are used to extract features. On 12000 numeral approach was tested and

reported 92.8% of recognition rate.

Marathi

G.G. Rajput and S.M. Mali [45] have shown use of Fourier descriptor and normalized

chain code to recognize isolated Marathi handwritten numerals. For classification

Support vector machine (SVM) is used on database of 12690 samples using fivefold

cross validation technique and have achieved 98.15% of recognition accuracy.

Shushma shelke and Shaila apte [46] have proposed work to recognize unconstrained

handwritten Marathi compound characters. Initial stage of feature extraction is based

on structural features and classification done as per structural parameters into 24 classes

where as in final stage wavelet transform is used as feature extraction. Wavelet features

extracted are then applied to neural network for classification. With wavelet

approximation features authors have achieved 94.22% and with modified wavelet

features 96.23% recognition rate for testing samples.

In research work presented by Reena Bajaj et.al [47] to recognize handwritten Marathi

numerals, three different types of features such as density features, moment features

and descriptive component features are used. Three different neural classifiers have

been used for classification of the numerals. Finally, the outputs of the three classifiers

are combined using a connectionist scheme.[45]

Vijay Rahul pawar and Arun Gaikwad [48] have worked to recognize isolated Marathi

character and various structural and statistical features are extracted like end points,

Chapter 2

25

middle bar, loop, end bar, aspect ratio etc. Self-organizing map (SOM) approach for

classification is used and 93% accuracy is achieved.

Tamil

N. Shanthi and K. Duraiswamy[49] have presented their work on impact of image size

on performance outcome to recognize unconstrained handwritten Tamil characters

using Support Vector Machine (SVM). Authors have collected handwritten samples

from different writers and performed digitization and preprocessing steps on scanned

images. Image is divided into different zones and pixel density is calculated and taken

as a base for feature selection. Support vector machine is used as classifier. Three

different image size with resolution 32 X 32, 48 X 48 and 64 X 64 is taken for

comparing performance and concluded that 64 X 64 yields best result. Authors have

reported 87.4% of recognition rate by training SVM with 106 different characters.

In [21][50] zone based features are extracted to recognize Tamil handwritten numerals

using nearest neighbor and Support vector machine classification approach and have

reported success rate of 93.9% and 94.9%.

To recognize handwritten Tamil text Bhattacharya et.al. [21][51] have proposed

algorithm where unsupervised clustering method and a supervised classification

technique is used.

For recognizing Tamil handwritten characters Kohonen self organizing maps (KSOM)

is used with a fine-tuning method, that uses global features and have reported a

recognition accuracy of 86%.[52][21]

R.Jagdeesh Kanan et. al. [53] have proposed improved approach for recognizing Tamil

character using octal graph conversion method. Certain process carried out on input

image includes preprocessing, segmentation, normalization before features are

extracted. Here input letter is converted to octal graph by representing given character

as a node of a graph where each node has eight fields. Authors have reported 82% of

success rate.

Chapter 2

26

To recognize handwritten Tamil characters S. Hewavitharana and H. C. Fernando [54]

have proposed use of a two-stage classification approach, for a subset of the Tamil

alphabet and their first stage, an unknown character is pre-classified into one of the

three groups: core, ascending and descending characters, then, the second stage,

members of the pre-classified group are further analyzed using a statistical classifier for

final recognition.[55]

As per concept given by N. Santhi et. al. [56] it is process when applied peels off a

pattern as may pixels as possible in such a way so that general shape of the patter is not

affected. Authors have used Hilditch’s algorithm for skeletonization and used otsu’s

global thresholding technique based on histogram for thresholding. Further authors

have used horizontal histogram profile for line segmentation and vertical histogram

profile used to segment character.

R. Jagadeesh Kannan and R. Prabhakar [55] have proposed discrete Hidden Markov

Model (HMM) to recognize offline cursive handwritten Tamil characters.

Preprocessing operations such as binarization, noise removal, skew correction is

performed and line, words and characters are segmented.

L.Anlo Safi and K.G.Srinivasagan [57] have shown use of zone based hybrid feature

extraction technique to recognize offline Tamil handwritten character. They have

extracted certain features from character image that includes number of horizontal line,

vertical line, diagonal lines with their total length in each zone. For classification feed

forward back propagation neural network is used and have concluded to have 98% of

highest recognition accuracy.

Telugu

B.V.Dhandra et.al. [23] have proposed neural network approach along with four feature

selection methods i.e. maximum profile distances, fill hole density, water reservoir and

directional density features for recognizing Telugu handwritten numerals and have

achieved 99.6% recognition rate.

Chapter 2

27

P. Bhaskara Rao et. al. [58] have used Zernike moments as shape descriptors and

identified it as rotation invariant.

Manipuri

Chandan Jyoti kumar and Sanjib kumar kalita have presented their research work in

[59] to recognize handwritten numeral of Manipuri script. Various feature extraction

technique used which includes background directional distribution (BDD), zone based

diagonal, projection histogram and histogram oriented gradient features. Support vector

machine (SVM) with five-fold cross validation with RBF kernel is used for

classification and testing.

For recognizing Manipuri script handwritten character T.Thokchom et.al [60] have

investigated a back propagation neural network approach. Probabilistic and fuzzy

features are extracted from character matrix. Network was trained and tested for

performance. They found recognition accuracy of 90.3%. [59]

Kannada

T V Ashwin et. al.[61] have segmented image into various zones to capture shape of

the characters in the angular directions. For classification Support vector machines

(SVM) is used and recognition rate obtained is 87% to 95%.

In [50] zone based features are extracted to recognize Kannada handwritten numerals

using nearest neighbor and Support vector machine classification approach and have

reported success rate of 97.75% and 98.2%. Further [62] authors has proposed zone and

image centroid based features extraction method is proposed. The image is divided into

equal zones, centroid of character is computed, average distance of each pixel to this

centroid point is calculated, and process is repeated for entire image.

In work [63] to recognize Kannada numerals zoning and direction chain code approach

is presented.

G.G.Rajput et.al.[64] have proposed a work to recognize printed and handwritten

Kannada numerals where all images are normalized to 40 x 40 followed by tracing

Chapter 2

28

boundary, determining chain code and 10 dimensional fourier descriptor are computed

to form feature vector. For classification multi-class Support vector machine (SVM) is

used on 2500 handwritten images with experiment carried out on five-fold cross

validation method and have reported average recognition rate of 97.34% for Kannada

numerals.

Directional features extraction technique i.e. histograms of direction chain code of the

contour points for recognition of handwritten Kannada numerals is proposed by

Mamatha H.R. et.al [65] where K-means clustering is used for classification. Further in

[66] an effort to recognize Kannada numerals have proposed fan beam projection a

variation of Radon transform is proposed as feature extraction method. 55 diverging

beams are considered by taking projection at by rotating source image at different

angles around center pixel. Size of feature vector for one numeral is 1 x 360. K-nearest

neighbor approach is proposed for classification and experiment is carried out on 1000

images and have achieved 86.29% recognition rate.

In research work presented [67] to recognize handwritten Kannada numerals two

feature extraction method is proposed i.e. Run length count (RLC) and directional chain

code is proposed and for classification K-nearest neighbor and linear classifier is used.

They have concluded to have higher recognition accuracy with directional chain code

against run length count.

Niranjan s.k. et.al. [68] have shown use of Fisher linear discriminant analysis (FLD)

for feature extraction as two dimension FLD and diagonal FLD. Distance measure

techniques such as correlation coefficient, Manhattan, Mahalanobis distance between

normed vectors, Mahalanobis, Euclidean, Minkowski, Modified manhattan, Modified

sq. Euclidean, Mean sq. error, Sq. Euclidean, weight angle, weight manhattan, weight

modified manhattan, weight modified SSE, weight SSE, Canberra and angle is used for

recognizing unconstrained Kannada handwritten characters.

Work proposed in [8] to recognize segmented or cursive characters modified direction

feature (MDF) extraction technique is proposed which is based on direction feature

technique. Extension to directional feature technique to integrate directional

information with finding transition between foreground and background pixels.

Chapter 2

29

Authors have tested MDF approach with neural network classifier and further compared

it with directional feature and transition features and concluded that MDF is

outperforming both. Proposed approach was tested on CEDAR database with

recognition accuracy above 89% is observed by authors.

Shivanand Rumma et.al. [69] have proposed use of radial basis function for recognizing

handwritten Kannada numerals.

Oriya

In a research work presented by Chaudhari et al.[70] water reservoir approach is used

for feature extraction and classification upon preprocessed image. Preprocessing steps

applied on handwritten image includes skew correction, segmentation of line, zone

detection, segmentation of word and character etc. and with proposed approach they

have achieved 96.3% of accuracy.

Gujarati

Character

In a work presented by Jayshree R. Prasad et.al. [71] pattern matching approach is

proposed to recognize offline handwritten character of Gujarati script. Various

preprocessing steps carried out on given scanned image as input png file which includes

removing salt and pepper noise, binary image, thinning, segmentation. Feed forward

neural network with back propagation learning is considered to train the network.

Template matching is used along with Cross correlation function. With this approach

authors have concluded to have average overall recognition rate of 71.66%.

Chayya Patel et.al. [72] have proposed research work to segment handwritten Gujarati

text lines into words. Morphological operations and projection profile based algorithm

was tested on 500 text line where some lines are repeated by same writer or different

writer and concluded to have batter accuracy rate is achieved in a case if text line is

written by same writer than different writers. Further segmentation approach using

vertical projection profile where all columns are checked where if ON pixels are less

then certain threshold value then it is considered as white space and separation is made

to extract words from character. In [73] work is reported to extract upper modifiers and

Chapter 2

30

lower modifiers for detecting upper zone and lower zone based on which middle zone

is identified. Input handwritten image is converted to binary image and Euclidean

distance transform is generated for to determine peak and valley for zones. Authors

have tested proposed work on 250 Gujarati words.

To recognize handwritten Gujarati character [74] proposes radial histogram approach

for feature extraction and Euclidean distance classifier is used as classification. Some

preprocessing steps applied includes binarization and normalization and 72 features

derived on the basis of radial histogram on handwritten input images and reported

26.86% of accuracy rate.

Number

Avani R. Vasant et. al.[75] have evaluated performance of different image size to

recognize offline handwritten Gujarati digits. Data samples collected and digitized at a

resolution of 150 dpi and performed various preprocessing operations. Images are

resized in three different sizes i.e. 7 x 5, 14 x 10 and 16 x 16. Back propagation neural

network is proposed for classification. Research approach was applied on collected

3900 handwritten digits where 2100 are used for testing and remaining for training

purpose and reported recognition rate of 87.29%, 88.52% and 88.76% on 7 x 5, 14 x 10

and 16 x 16 image sizes and concluded that with 16 x 16 image size best recognition

rate is achieved as more details can be captured if image size increase.

In a work presented in [76] to recognize handwritten Gujarati numeral support vector

machine (SVM) approach is proposed. Handwritten samples digitized at 300 dpi and

preprocessing steps were applied to remove noise, skew correction, enhancing image

by morphological operations, size normalization. All images were normalized as 40 x

40 for which nearest neighbor interpolation technique is applied, then image is

converted to binary then image is skeletonized to prepare it for next phase of

recognition. For feature extraction image is divide into various size boxes and based on

that four feature sets are derived from which classification is carried out using SVM.

Authors have concluded to have 90.55% of average recognition rate in which ‘0’ is

having highest recognition rate with proposed approach.

Chapter 2

31

Bhaeti M.J. et. al. [77] have compared two classifiers namely K-nearest neighbor

classifier (KNN) and Principal component analysis (PCA) to recognize offline

handwritten Gujarati numerals. Affine moment invariant (AMI) model is used for

feature extraction and concluded that KNN is better classifier than PCA as recognition

rate reported is 90.04% against 84.1% respectively. In [78] authors have extended

comparison with Principal component analysis (PCA), Support vector machine (SVM),

K-nearest neighbor (KNN) and Gussian distribution function along with Affine

moment invariant (AMI) as feature extraction method and achieved recognition rate for

identifying Gujarati digits as 84.1%, 92.28%, 90.04% and 87.2% respectively with

conclusion that SVM is better classifier.

Kamal moro et. al.[79] have achieved 80.5% of overall performance and in their

research work to recognize Gujarati handwritten numeral using neural network

classifier.

A.A. Desai [80] has collected 300 samples of 300dpi with initial size of each numeral

as 90x90. the author then adjusted the contrast by CLAHE i.e. contrast limited adaptive

histogram equalization algorithm considering 8x8 tiles and 0.01 as contrast

enhancement constant. The boundaries are then smoothed out by median filter of 3x3

neighborhoods. Image is then reconstructed to the size of 16x16 pixels using nearest

neighbor interpolation. For feature extraction four profile vectors are used as an

abstracted feature of identification of digit. Five more patterns for each digit are created

in both clockwise and anticlockwise directions with the difference of 2degrees each up

to 10◦. A feed forward back propagation neural network is used for Gujarati numeral

classification with 278 sets of various digits. Out of these 278 sets, 11 sets were created

by a standard font. From the 265 sets the author recorded the success rate for standard

fonts as 71.82%, for handwritten training sets as 91.0% while for testing sets as a score

of 81.5% was recorded.

Chapter 2

32

REFERENCES

[1] D. Kumar Patel, “Handwritten Character Recognition Using Multiresolution

Technique and Euclidean Distance Metric,” J. Signal Inf. Process., vol. 03, no.

02, pp. 208–214, 2012.

[2] R. Sheth, N. C. Chauhan, M. M. Goyani, and K. A. Mehta, “Handwritten

Character Recognition System using Chain code and Correlation Coefficient,”

Communication, pp. 31–36, 2011.

[3] “Handwritten English Character Recognition Using Neural Networks”, Anita

Pal and Dayashanker Singh,vol 1,no 2, July-December 2010.

[4] P. Singh, “Radial Basis Function For Handwritten Devanagari Numeral

Recognition,” Int. J., vol. 2, no. 5, pp. 126–129, 2011

[5] S. Saha, W. Benga, and N. Paul, “Optical Character Recognition using 40-point

Feature Extraction and Artificial Neural Network,” vol. 3, no. 4, pp. 495–502,

2013.

[6] B. Verma, M. Blumenstein, and M. Ghosh, “A novel approach for structural

feature extraction Contour vs. direction, vol. 25, pp. 975–988, 2004.

[7] H. Yamada, Y. Nakano, Cursive handwritten word recognition using multiple

segmentation determined by contour analysis, IEICE Trans. Inform. Syst. E79-

D (1996) 464–470

[8] M. Blumenstein, X. Yu, and B. Verma, “An investigation of the modified

direction feature for cursive character recognition,” vol. 40, pp. 376–388, 2007

[9] F. Kimura, N. Kayahara, Y. Miyake, M. Shridhar, Machine and human

recognition of segmented characters from handwritten words, Proceedings of

Fourth International Conference on Document Analysis and Recognition, 1997,

pp. 866–869.

[10] S. Singh, M. Hewitt, Cursive digit and character recognition on cedar database,

Int. Conf. Pattern Recognition (2000) 569–572.

[11] [online] building_skeletal_graphsfor_.pdf.

[12] C. Liu, K. Nakashima, H. Sako, and H. Fujisawa, “Handwritten digit

recognition?: investigation of normalization and feature extraction techniques,”

vol. 37, pp. 265–279, 2004.

Chapter 2

33

[13] R. M. O. Cruz, G. D. C. Cavalcanti, and T. I. Ren, “Handwritten Digit

Recognition Using Multiple Feature Extraction Techniques and Classifier

Ensemble,” 2010.

[14] L. Heutte, T. Paquet, J. V. Moreau, Y. Lecourtier and C. Olivier, “A

structural/statistical feature based vector for handwritten character recognition”,

Pattern Recognition Letters, Vol. 19(7), pp. 629-641, 1998.

[15] N. F. Abubacker, “An Approach for Structural Feature Extraction for Distorted

Tamil Character Recognition,” vol. 22, no. 4, pp. 24–28, 2011.

[16] L. L. Lee and N. R. Gomes, “Disconnected handwritten numeral image

recognition”, in the Proceedings of 4th ICDAR, pp. 467-470, 1997.

[17] C. Science and S. Publications, “Recognition of Hand Printed Characters Based

on Simple Geometric Features Sami M . Halawani and Ibrahim A . Albidewi

Department of Computer Science , Faculty of Computing and Information

Technology , King Abdulaziz University , Jeddah , Kingdom Saudi Arabia,”

vol. 6, no. 12, pp. 1518–1523, 2010.

[18] M. Karic and G. Martinovic, “Improving Offline Handwritten Digit

Recognition Using Concavity-Based Features Feature Extraction,” vol. 8, no. 2,

pp. 220–234, 2013

[19] V. V. Kumar, A. Srikrishna, and B. R. Babu, “Classification and recognition of

handwritten digits by using mathematical morphology,” vol. 35, no. August, pp.

419–426, 2010

[20] Lakshmi, C.V., Jain, R., Patvardhan, C. (2007) “Handwritten cybernetics, and

informatics, Hyderabad, pgs.654-659 Devnagari Numerals Recognition with

Higher Accuracy” In International Conference on Computational Intelligence

and 2007 Multimedia Applications: Volume : 3 : 255

[21] Baheti M.J., Kale K.V., “Gujarati Numeral Recognition Affine Invariants

Moment Approach”, 14th International Conference on Recent Trends in

Engineering and Technology, May-2012, IJCECSCSE

[22] Pal, U., Wakabayashi, T., Kimura, F. 2009. “Comparative Study of Devanagari

Handwritten Character Recognition Using Different Feature and Classifiers”

10th International Conference on Document Analysis and Recognition, ICDAR

'09 : 1111

Chapter 2

34

[23] B. V Dhandra, “Kannada , Telugu and Devanagari Handwritten Numeral

Recognition with Probabilistic Neural Network?: A Novel Approach,”

Architecture, pp. 83–88, 2010

[24] M. Jangid, K. Singh, and R. Rani, “Performance Comparison of Devanagari

Handwritten Numerals Recognition,” Int. J., vol. 22, no. 1, pp. 1–6, 2011.

[25] M. Jangid, R. Dhir, R. Rani, and K. Singh, “SVM Classifier for Recognition of

Handwritten Devanagari Numeral,” Processing, no. Iciip, 2011.

[26] S. Arora, “Recognition of Non-Compound Handwritten Devnagari Characters

using a Combination of MLP and Minimum Edit Distance,” J. Comput. Sci.,

no. 4, pp. 107–120.

[27] U. Pal, N. Sharma, T. Wakabayashi and F. Kimura, “Off-Line Handwritten

Character Recognition of Devnagari Script”, In Proc. 9th International

Conference on Document Analysis and Recognition, 2007, pp. 496-500.

[28] U. Pal, T. Wakabayashi, N. Sharma and F. Kimura, “Handwritten Numeral

Recognition of Six Popular Indian Scripts”, In Proc. 9th International

Conference on Document Analysis and Recognition, 2007, pp. 749-753.

[29] D. B. T. RAJIV KAPOOR, "Skew angle detection of a cursive handwritten

Devanagari script Character Image," J. Indian Inst. Sci, pp. 161-175, May-Aug

2002

[30] R. a. S.C.Mehrotra, "Recognition Handwritten Devanagari Numerals,"

International journal of Computer processing of Oriental languages, 2008

[31] S. K. a. C. Singh, "A Study of Zernike Moments and its use in Devnagari

Handwritten Character Recognition," in Intl.Conf. on Cognition and

Recognition, 2005

[32] S. K. Shrivastava and S. S. Gharde, “Support Vector Machine for Handwritten

Devanagari Numeral Recognition,” vol. 7, no. 11, pp. 9–14, 2010.

[33] Prachi Mukherji, Priti Rege, “Shape Feature and Fuzzy Logic Based Offline

Devnagari Handwritten Optical Character Recognition”, Journal of Pattern

Recognition Research 4,pp. 52-68, 2009

[34] K. S. Siddharth and R. Rani, “HANDWRITTEN GURMUKHI CHARACTER

RECOGNITION,” Technology, vol. 144011

[35] P.M. Patil, T.R. Sontakke, “Rotation, scale and translation invariant handwritten

Devanagari numeral character recognition using general fuzzy neural network”,

Pattern Recognition, Elsevier, Vol. 40, Issue 7, pp. 2110-2117, July 2007

Chapter 2

35

[36] A. N. Holambe, “Combining Multiple Feature Extraction Technique and

Classifiers for Increasing Accuracy for Devanagari OCR,” no. 4, pp. 38–41,

2013.

[37] A. N. Holambe, D. R. C. Thool, and D. S. M. Jagade, “Printed and Handwritten

Character & Number Recognition of Devanagari Script using gradient

features,” Int. J. Comput. Appl., vol. 2, no. 9, pp. 38–41, Jun. 2010.

[38] Dinesh Acharya U, N V Subba Reddy and Krishnamoorthi, “Isolated

handwritten Kannada numeral recognition using structural feature and K-means

cluster”, IISN-2007, pp-125- 129, 2007.

[39] Sharma D. and Jhajj P. (2010) “Recognition of Isolated Handwritten Characters

in Gurmukhi Script” International Journal of Computer Applications 4(8):9-17.

[40] Singh K. S. Jangid M., Dhir R. Rajneesh Rani,(2011) “Handwritten Gurmukhi

Character Recognition Using Statistical and Background Directional

Distribution Features” International Journal on Computer Science and

Engineering 3(6): 2332-2345.

[41] R. D. R. R. Kartar Singh Siddharth, "Handwritten Gurmukhi Numeral

Recognition using Different Feature Sets," International Journal of Computer

Applications, vol. 28, no. 2, August 2011.

[42] S. Singh, A. Aggarwal, and R. Dhir, “Use of Gabor Filters for Recognition of

Handwritten Gurmukhi Character,” vol. 2, no. 5, pp. 234–240, 2012.

[43] A. S. Mashiyat, “Bangla off-line Handwritten Character Recognition Using

Superimposed Matrices,” Engineering.

[44] G. Read, “A System for Bangla Handwritten Numeral Recognition.”

[45] G. G. Rajput and S. M. Mali. Fourier descriptor based Isolated Marathi

Handwritten numeral Recognition. International journal of Computer

Application.2010, pp. 5120/724-1017.

[46] S. Shelke, “A Multistage Handwritten Marathi Compound Character

Recognition Scheme using Neural Networks and Wavelet Features,” vol. 4, no.

1, pp. 81–94, 2011.

[47] Reena Bajaj, Lipika Dey and Santanu Chaudhuri 2002. Marathi numeral

recognition by combining decision of multiple connectionist classifiers, Vol. 27,

Part 1,59–72, February 2002.

[48] V. R. Pawar, “Performance Evalution of Multistage Offline Marathi Script

Recognition System,” vol. 88, no. 4, pp. 33–39, 2014.

Chapter 2

36

[49] C. Science and S. Publications, “Performance Comparison of Different Image

Sizes for Recognizing Unconstrained Handwritten Tamil Characters using

SVM N . Shanthi and 2 K . Duraiswamy Department of Information

Technology, Department of CSE , K . S . Rangasamy College of Technology,

Tiruchengode , India,” vol. 3, no. 9, pp. 760–764, 2007.

[50] Rajashekararadhya, S.V. Vanaja Ranjan, P. (2009) “Zone-based hybrid feature

extraction algorithm for handwritten numeral recognition of two popular Indian

scripts”, Nature & Biologically Inspired Computing, p 526 – 530

[51] Bhattacharya, U. Ghosh, S.K. Parui, S.K. (2007) “A Two Stage Recognition

Scheme for Handwritten Tamil Characters” Proc. ICDAR, 1:511 – 515.

[52] Sarveswaran, K. Ratnaweera, D.A.A.C (2007) “An adaptive technique for

handwritten Tamil character recognition”, Proc. ICIAS, p151 – 156

[53] R. P. R. Jagdeesh Kannan, “An Improved Handwritten Tamil Character

Recognition System using Octal Graph” J. Comput. Sci., vol. 4, no. 7, pp. 509–

516, 2008.

[54] Hewavitharana, S, and H.C. Fernando, 2002. A Two Stage Classification

Approach to Tamil Handwriting Recognition, pp: 118-124, Tamil Internet 2002,

California, USA.

[55] R. J. Kannan, “Off-Line Cursive Handwritten Tamil Character Recognition,”

Signal Processing, vol. 4, no. 6, pp. 351–360, 2008.

[56] K. D. N Shanthi, "Preprocessing Algorithms for the Recognition of Tamil

Handwritten Characters," cochin, 2005

[57] L. A. Safi and P. G. Scholar, “Offline Tamil Handwritten Character Recognition

using Zone based Hybrid Feature Extraction Technique,” vol. 65, no. 1, pp. 12–

16, 2013.

[58] P. B. Rao, “Feature Extraction Using Zernike Moments,” vol. 2, no. 2, pp. 228–

234.

[59] C. J. Kumar, “Recognition of Handwritten Numerals of Manipuri Script,” vol.

84, no. 17, pp. 1–5, 2013.

[60] Tangkeshwar Thokchom, P.K.Bansal, Renu Vig and Seema Bawa,

“Recognition of Handwritten Character of Manipuri Script”, Journal Of

Computers, Vol. 5, No. 10, Oct 2010.

Chapter 2

37

[61] T V Ashwin and P S Sastry “A font and size independent OCR system for

printed Kannada documents using support vector machine”. Saadhanaa, Vol.

27, Part 1, 2002, pp. 35–58.

[62] S. R. a. P. V. Ranjan, "Isolated handwritten Kannada digit recognition: A novel

approach," in Proceedings of the International Conference on Cognition and

Recognition, 2008.

[63] N. Sharma, U. Pal, F. Kimura, "Recognition of Handwritten Kannada

Numerals", ICIT, pp. 133-136, ICIT'06, 2006.

[64] G. G. Rajput, R. Horakeri, and S. Chandrakant, “Printed and Handwritten

Mixed Kannada Numerals Recognition Using SVM,” Int. J., vol. 02, no. 05, pp.

1622–1626, 2010.

[65] S. M. K. V. A. V. P. S. V. L. M. Mamatha H.R, "Recognition of Handwritten

Kannada Numerals using Directional Features and K-Means," in International

Conference on Computational Intelligence and Communication Systems, 2011.

[66] H. R. Mamatha, S. M. K, V. G. Raj, and S. S. Jois, “Fan Beam Projection Based

Features to Recognize Handwritten Kannada Numerals,” Computer (Long.

Beach. Calif)., vol. 9, pp. 257–262, 2011.

[67] S. Murthy, “Feature Based Recognition of Handwritten Kannada Numerals – A

Comparative Study,” Communication.

[68] S. K. Niranjan, V. Kumar, H. K. G, and M. A. V N, “FLD based Unconstrained

Handwritten Kannada Character Recognition,” Int. J., vol. 2, no. 3, pp. 21–26,

2009.

[69] S. Rumma, “Handwritten Kannada Numeral Recognition using Radial Basis

Function,” vol. 98, no. 8, pp. 18–20, 2014.

[70] B. B. Chaudhari, U. Pal and M. Mitra “Automatic recognition of printed Oriya

script” Sadhana Vol 27(1) 200) pp23-34.

[71] J. R. Prasad, “Offline Handwritten Character Recognition of Gujrati Script

using Pattern Matching,” Comput. Eng., 2003.

[72] C. Patel and A. Desai, “Segmentation of Text Lines into Words for Gujarati

Handwritten Text,” Image Process., pp. 130–134, 2010.

[73] C. Patel, “Zone Identification for Gujarati Handwritten Word,” 2011.

[74] L. Shah, R. Patel, S. Patel, and J. Maniar, “Handwritten Character Recognition

using Radial Histogram,” vol. 2, no. 4, pp. 24–28, 2014.

Chapter 2

38

[75] A. R. Vasant, “Performance Evaluation of Different Image Sizes for

Recognizing Offline Handwritten Gujarati Digits using Neural Network

Approach,” in 2012 International Conference on Communication Systems and

Network Technologies, 2012, pp. 271–274.

[76] M. Maloo, K. V Kale, and I. Technology, “SUPPORT VECTOR MACHINE

BASED GUJARATI NUMERAL,” Int. J., vol. 3, no. 7, pp. 2595–2600, 2011.

[77] M. J. Baheti, K. V Kale, and M. E. Jadhav, “COMPARISON OF

CLASSIFIERS FOR GUJARATI NUMERAL RECOGNITION,” Int. J. Mach.

Intell., vol. 3, no. 3, pp. 93–96, 2011.

[78] M. J. Baheti, “Gujarati Numeral Recognition: Affine Invariant Moments

Approach,” Soft Comput., pp. 140–146, 2012.

[79] M. f. Kamal moro, "Gujarati Handwritten Numeral Optical Character through

neural network and skeletonization," jurnal of sistem komputer, vol. 3, no. 1,

pp. 40-43, 2013

[80] A. a. Desai, “Gujarati handwritten numeral optical character reorganization

through neural network,” Pattern Recognit., vol. 43, no. 7, pp. 2582–2589, Jul.

2010.