offline marathi handwritten character recognition … marathi handwritten character recognition...

Offline Marathi Handwritten Character Recognition Using SVM Classifier

Based on Cloud Computing: Review

Sanjay S. Gharde Leena Y.Rane

Asst. Professor M.E.( CSE) IInd Year

SSBT COET, Jalgaon SSBT COET, Jalgaon

[email protected] [email protected]

Mob:-9422344964

Mob:-9405705760

Abstract

Handwritten recognition has been one of the active and challenging research areas in the field of image

processing. However, most of the current work in

these areas is limited to English and a few oriental

languages with single computer. Many systems and

classification algorithms have been proposed in the

past years on handwritten characters recognition in

various languages like English, Persian, Arabian and

Devanagari scripts also. Researchers had been

worked on Handwritten Devanagari characters by

applying different techniques on single PC but using

client and server recognize the Marathi handwritten

character based on cloud computing is challenging.

Previously researchers had been worked on online

Chinese characters based on cloud computing. So,

this research work has been conducted on offline

Marathi Handwritten character based on cloud

computing and this system is very important because person can access any data from anywhere. There

are various feature extraction techniques and

classifiers for recognition of Marathi characters but

Freeman chain code histogram and Gradient feature

extraction techniques are better because freeman

chain code approach is robustness to small variation

and easy to implement and the gradient technique

can be easily used to gray scale images and are

robust against image noise and edge direction

fluctuation. SVM classifier is better than KNN and

ANN classifier because of its complexity of training,

flexibility, classification accuracy and complexity

which is given bellow.

Keywords Devnagari Marathi Character Recognition, Off-line

Handwriting Recognition, Pre-processor, Feature Extraction, Image Classification.

1. Introduction

Offline handwriting recognition is one of the

challenging problems in pattern recognition. The

main challenges are wide variety of handwriting

styles, large varieties of pen-type, poor image quality

and a lack of ordering information of strokes. In the

this work a cloud-based recognition architecture is

presented. It has possible to be useful not only to end users, but also to researchers in the field. A cloud

infrastructure can support in the capture of

recognition history[1]. Additionally, such a model

has a number of other advantages: First, it allows the

writer to train the model only once and then use the

cloud with any device connected to the Internet.

Secondly, it gives the user access to various default

collections of training samples across different

alphabets, languages, and domains. Thirdly, it

provides a higher level of control over the

classification results and correction history. Further,

Cloud computing platform differentiate from the

traditional client server model. Cloud computing

offers the same super computer services over the

Internet and dynamically allocates distributed

computing resources. Cloud server can easily handle

service requests of multiple users simultaneously; Cloud computing system supports cross-platform

client terminals. Therefore, it would be more suitable

and easier for recognition the offline Marathi

handwritten character recognition[2].

2. Related Work

The first research work report on

handwritten Devnagari characters was published in

1977[3]. For Indian languages most of research work

is performed on firstly on Devnagari script and

secondly on Bangla script. U. Pal and B.B.

Chaudhury[4] presented a survey on Indian Script

Character Recognition. This

Leena Y Rane et al , International Journal of Computer Science & Communication Networks,Vol 4(3),76-83

76

ISSN:2249-5789

This paer introduce recognition of many Indian

language scripts including Devnagari, Bangla, Tamil,

Oriya, Gurumukhi, Gujarati and Kannada.Bikash

Shaw, Swapan Kr. Parui, Malayappan Shridhar[5]

used a novel segmentation based approach for

recognition of offline handwritten Devanagariwords.

They also used Stroke based features for feature

vectors.In this paper Hidden Markov model is used

for recognition at pseudocharacter level.Sandhya Arora et al.[6] used intersection, shadow features,

chain code histogram and straight line fitting feature

extraction techniques and weighted majority voting

technique for combining the classification decision

obtained from different Multi-Layer Perceptron

(MLP) based classifier. They obtained 92.80%

accuracy results for handwritten Devnagari

recognition. Prachi Mukherji and Priti Rege[7]used

shape features and fuzzy logic to recognize offline

Devnagari character recognition. They segmented the

character into strokes using structural features like

endpoint, cross-point, junction points, and thinning.

Using tree and fuzzy classifier they classified the

segmented shapes or strokes as left curve, right

curve, horizontal stroke, vertical stroke, slanted lines

etc. They obtained average 86.4% accuracy.U. Pal,

Wakabayashi and Kimura also presented comparative study of Devnagari handwritten character recognition

using different features and classifiers[8]. They used

four sets of features based on curvature and gradient

feature extraction from binary as well as gray scale

images and compared results using 12 different

classifiers as concluded the best results 94.94% and

95.19% for features extracted from binary and gray

image respectively obtained with Mirror Image

Learning (MIL) classifier. They also accomplished

curvature features to use for better results than

gradient features for most of classifiers. Mallikarjun

Hangarge[9] investigate a texture for determining the

script of handwritten document image, Further, K

nearest neighbour classifier is used to classify 300

text blocks as well as 400 text lines into one of the

three major Indian scripts: English, Devnagari and

Urdu, Using morphological filters 13 spatial spread features are extracted. K. Roy, S. Kundu Das, Sk Md

Obaidullah[10] uses very simple and efficient

feature at component level. In this paper fractal-based

features, component based feature and Topological

features, series of classifiers were used. Accuracy of

the system is at present 89.48% on the test set

without rejection. In this approach gradient features

using Yan Gao, Lanwen Jin, Cong He, Guibin

Zhou[2] are extracted and then these features are

converted into 8 directional features. In chain code

histogram to character contour there are basically two

ways to determine the direction codes of contour

pixels. In one way, the order of contour pixels is

obtained by contour tracing, and then the chain codes

are calculated. And secondly we use the gradient

representation as the basis for extraction of features.

These algorithms require a few simple arithmetic

operations per image pixel which makes them

suitable for real-time applications. Ved Prakash

Agnihotri used Diagonal based feature extraction for

the handwritten Devanagari script. After that these

feature of each character image is converted into

chromosome bit string of length 378. In recognition

step using fitness function in which find the

Chromosome difference between unknown character and Chromosome which are store in data base[11].

Pankaj Kumawat , Asha Khatri , Baluram

Nagaria[12] used the Curve let transform and

Invariant moments for character recognition problem.

They compare the performance of HMM based

technique with combined HMM- SVM based

technique and found that for some combined HMM-

SVM technique is better than HMM. Combined

HMM-SVM classifier improve the problem of HMM

classifier of multiple detection of Class.

After describing cloud computing in Section

3 Handwritten character recognition techniques is

discussed in Section 4, Segmentation is discussed in

section 5, and Feature Extraction types discussed in

section 6, Character Classification is discussed in

Section 7. Proposed system of Cloud based

Handwritten Character Recognition is discussed in Section 8.Finally, Discussion on review is discussed

in Section 9.

3. Cloud Computing

Cloud computing is the junction of several

concepts from resource pooling, virtualization,

dynamic provisioning, and utility computing.

Applications are easily deployed where the

underlying technology components can expand and

contract with the flow of the business cycle. Cloud

computing is a new method to add capability to a

computer without licensing new software, investing

in new hardware or infrastructure or training new personnel. Applications are purchased and run over

the network listed users’ desktop. A cloud server is a

logical server that is built, hosted and delivered

through a cloud computing platform over the

Internet. Cloud servers possess and exhibit similar

capabilities and functionality to a typical server but

are accessed remotely from a cloud service provider.

Most of the current clouds are built on top of modern

data centers. It incorporates Infrastructure as a

Service (IaaS), Platform as a Service (PaaS), and

Software as a Service (SaaS), and provides these

services like utilities, so the end users are billed by

how much they used[13]. IaaS provides service such

as memory, CPU and storage. It reduces hardware

cost. That means customer use a virtualized server

and running software on it. Amazon EC2, is the best

example of IaaS for storage and maintaining virtual servers.


77

ISSN:2249-5789

4. Handwritten Character Recognition

Techniques The general steps of handwriting recognition

include preprocessing, segmentation, feature extraction, classification and so on. The

preprocessing technology consists of binarization,

normalization, and smoothing and slant correction.

The widely used feature extraction methods for

handwritten Marathi character recognition include 8-

directional feature extraction and Gradient feature

extraction etc. In classification step, Support Vector

Machine (SVM) classifier is mainly used.

5. Segmentation

Segmentation is the most important part of

the pre-processing stage. It allows the recognizer to

extract features from each individual character. In the

more complicated case of handwritten text, the

segmentation problem becomes much more difficult as

quality have a tendency to be connected to each other,

overlapped or unclear. Segmentation is done to break

the single text line, single word and single character

from the input document. For isolated characters or

numerals, segmentation task is not that difficult.

However, for joint and complex strings more advanced

techniques required to be employed. There are two

types of segmentations: 1. External segmentation isolates various writing units such as paragraphs,

sentences or words. 2. Internal segmentation, which is

isolation of letters. Character segmentation strategies

are divided into three techniques: Explicit

Segmentation-This approach is used to identify the

smallest possible word segments that may be smaller

than letters, but surely cannot be segmented further. It

is done by the process of analysis. Implicit

Segmentation- This is based on recognition. It

searches the image for components that match

predefined classes. Mixed Strategies- They combine

explicit and implicit segmentation in a hybrid way[14].

6. Feature Extraction Techniques

After the segmentation procedure the feature

set is extracted. Extracted features from the character

images are used to train the Support vector Machine.

In this stage, the features of the characters that are

used for classifying them at recognition stage are

extracted. This is an important stage as its efficient

performance improves the recognition rate and

reduces the misclassification. A good feature set

should represent characteristic of a class that help

distinguish it from other classes. The widely used

feature extraction methods are Template matching,

Deformable templates, Unitary Image transforms,

Graph description, Projection Histograms, Contour

profiles, Zoning, Geometric moment invariants,

Zernike Moments, Spline curve approximation,

Fourier descriptors, Gabor feature. Due to the nature

of handwriting with its high degree of inconsistency

and ambiguity extracting these features, is a difficult

task. Chain code Histogram of Character contour and

gradient feature extraction techniques are used in this paper[15].

6.1 Chain Code Histogram of Character

Contour

Given a scaled binary image, first find the

contour points of the character image. Consider a 3 × 3 window surrounded by the object points of the

image. If any of the 4-connected neighbor points is a

background point then the object point (P), as shown

in Figure 1 is considered as contour point.

Figure 1. Contour point detection

The contour following procedure uses a contour

representation called “chain coding” that is used for

contour shown in Figure 1.4. Each pixel of the

contour is assigned a different code that indicates the

direction of the next pixel that belongs to the contour

in some given direction. Chain code provides the points in relative position to one another, independent

of the coordinate system. In this methodology of

using a chain coding of connecting neighboring

contour pixels, the points and the outline coding are

captured. Contour following procedure may proceed

in clockwise or in counter clockwise direction. Here,

we have chosen to proceed in a clockwise direction

[16].

(a) (b) (c)

Figure 2. Chain Coding: (a) direction of

connectivity, (b) 4-connectivity, (c) 8-

connectivity. Generate the chain code by

detecting the direction of the next-in-line

pixel

6.2 Gradient Feature Extraction

The gradient measures the magnitude and

direction of the greatest change in intensity in a small


78

ISSN:2249-5789

neighborhood of each pixel. (In what follows,

"gradient" refers to both the gradient magnitude and

direction). Gradients are computed by means of the

Sobel operator. The Sobel templates used to compute

the horizontal (X) & vertical (Y) components of the

gradient are shown in Figure 3. The arc tangent of

the Gradient and Gaussian filter. A modified

quadratic classifier is applied on the features of

handwritten characters for recognition. Elastic matching (EM) technique based on an Eigen

deformation (ED) for recognition of handwritten

Devnagari characters is proposed in [17].

Figure 3. Contour point detection

Horizontal Component Vertical Component

Figure 3 Sobel masks for gradient Given an input

image of size D1×D2, each pixel neighborhood is

convolved with these templates to determine these X and Y components, Sx and Sy, respectively. Eq. (1)

and (2) represents their mathematical representation:

(1) (2) Here, (i,j) range over the image rows (D1)

and columns (D2), respectively. The gradient

strength and direction can be computed from the

gradient vector [Sx, Sy] T as shown below using Eq.

(3) and (4) Sx i, j = I i − 1, j + 1 + 2 ∗ I i, j + 1 + I i + 1, j + 1 − i − 1, j − 1 − 2 ∗ I i, j − 1 − I i + 1, j − 1 (1)

𝑆𝑦 𝑖, 𝑗 = 𝐼 𝑖 − 1, 𝑗 − 1 + 2 ∗ 𝐼 𝑖 − 1, 𝑗 + 𝐼 𝑖 − 1, 𝑗 + 1 − 𝐼 𝑖 + 1, 𝑗 − 1 − 2 ∗ 𝐼(𝑖 + 1, 𝑗) − 𝐼 𝑖 + 1, 𝑗 + 1

(2)

The gradient magnitude is then calculated as:

𝑟 𝑖, 𝑗 = Sx2 i, j + Sy2 i, j (3)

Gradient direction is calculated as:

𝜃 𝑖, 𝑗 = 𝑡𝑎𝑛^(−1) 𝑆𝑦 𝑖 ,𝑗

𝑆𝑥 𝑖 ,𝑗 (4)

cos ∝ +𝑐𝑜𝑠𝛽

= 2 cos 1

2 𝛼 + 𝛽 cos[

1

2 𝛼 − 𝛽 ]

After obtaining gradient vector of each

pixel, the gradient image is decomposed into four

orientation planes or eight direction planes (chain

code directions) as shown in Figure 1.4.

Figure 4. Contour point detection of gradient vector

After this, gradient vector of each pixel is

decomposed into components along these standard

direction planes. If a gradient direction lies between

two standard directions, it is decomposed into two

components in the two standard directions.

7. Classifiers

The decision making part of a recognition

system is the classification stage and it uses the

features extracted from the previous stage. There are

various methods for classification. K Nearest

Neighbor (KNN), Artificial Neural Network(ANN)

and Support Vector Machine (SVM) In this paper,

we discuss the characteristics of the some

classification methods that have been successfully applied to Offline Devnagari Marathi character

recognition and results of SVM classification is better

than other classification methods, applied on

Handwritten Devnagari characters.

7.1 Support Vector Machine Classifier

Support Vector Machine is relatively new

classification technique producing efficient

recognition results. SVM is a new type of hyperplane

classifier developed based on statistical learning

theory given by Vapnik. Basically SVM is a binary

(two class) linear classifier in kernel induced feature

space and is formulated as a weighted combination of

kernel functions on training examples. The kernel

function represents the inner product of two vectors

in linear/nonlinear feature space. Multiclass

classification is accomplished by combining multiple

binary SVM classifiers. Support Vector Machine is

supervised learning tool which is used for

classification and regression. The basic SVM takes a

set of input data and predicts, for each given input. A

set of training examples given each marked as

belonging to one of two categories, a SVM training algorithm builds a model that assigns new examples

into one category or the other. A support vector

machine constructs a hyper plane or set of hyper

planes in a high dimensional space, which can be

used for classification, regression or other tasks.

Naturally, a good separation is achieved by the hyper

plane that has the largest distance to the nearest

training data point of any class (so-called functional

margin), since in general the larger the margin the

lower the generalization error of the classifier[18].

Figure 5. Feature transformation using SVM


79

ISSN:2249-5789

There are two types of SVM:1)Linear SVM

& 2) Non Linear SVM. Linear SVM is the newest

extremely fast machine learning (data mining)

algorithm for solving multiclass classification

problems from ultra large data sets that implements

an original proprietary version of a cutting plane

algorithm for designing a linear support vector

machine. Linear SVM is a linearly scalable routine

meaning that it creates an SVM model in a CPU time which scales linearly with the size of the training data

set. SVM models clearly show its superior

performance when high accuracy is required.

Features: 1. Efficiency in dealing with extra large data sets

(say, several millions training data pairs),

2. Solution of multiclass classification problems

with any number of classes,

3. Working with high dimensional data (thousands

of features, attributes) in both sparse and dense

format,

4. No need for expensive computing resources

(personal computer is a standard platform),

5. Ideal for contemporary applications in digital

advertisement, e-commerce, web page

categorization, text classification, bioinformatics,

proteomics, banking services and many other areas.

Linear functions can control the complexity of

a learning machine. To avoided the problem of

dealing with too complex functions at the value of

being able to solve only linearly separable problems.

Following diagram will show how to extend the

linear SVM for constructing a very rich set of non-

linear decision functions while at the same time

controlling their complexity.

Figure 6. Three different views on the same two class separation problem.(a) A linear

separation of the input points in not possible without errors.(b) A better separation is

provided by a non linear surface in the input space. (c) This non linear surface

corresponds to a linear surface in a feature space

7.2 Kernels of SVM Classifiers

In addition to basic linear kernel, many other

kernels are also proposed by researchers. Following

are the four basic kernels used in SVM

classifications:

Linear: 𝐾 𝑥𝑖, 𝑥𝑗 = 𝑥𝑖𝑇. 𝑥𝑗 Polynomial : 𝐾 𝑥𝑖, 𝑥𝑗 = (𝛾𝑥𝑖𝑇 .𝑥𝑗 + 𝑟)𝑑 . 𝛾 > 0. Radial Basis Function(RBF):

K xi, xj = exp −γ xi − xj 2 , γ > 0.

Sigmoid: K xi. xj = tanh γxiTxj + r . Here and are kernel parameters. Among these kernels

RBF kernel is mostly used[19].

Figure 7. Linear separating hyper planes for the separable case. The support vectors are

circled

𝐷 𝑥 = 𝑊 𝑡 𝑥 +b (1) where w is an m-dimensional vector, b is a scalar,

𝑦𝑖𝐷 𝑋𝑖 ≥ 1 − ξi𝑓𝑜𝑟 𝑖 = 1, … . , 𝑀 (2)

Here ξi are nonnegative slack variables. The distance

between the separating hyper plane D(x) = 0 and the training datum, with ξi = 0, nearest to the hyper plane

is called margin. The hyper plane D(x) = 0 with the

maximum margin is called optimal separating hyper

plane. To determine the optimal

separating hyper plane, we minimize

1/2| 𝑊 |2 +C ξi 𝑀𝑖−1 (3)

subject to the constraints:

𝑦𝑖 𝑊𝑡𝑋𝑖 + 𝑏 ≥ 1 − ξi for i = 1,… . , M (4) where, C is the margin parameter that determines the

tradeoff between the maximization of the margin and

minimization of the classification error. The data that

satisfy the equality in (4) are called support vectors.

[20].

8. Proposed System

8.1 Challenges solution of System 1) How to identify variability for same character?

The main challenges in offline handwritten

character for Indian languages is to build a system

that is able to distinguish between variation in

writing the same stroke and minor variation in

similar characters in the script. The recognition

system should be able to distinguish between

structural or shape variations across similar

characters and the natural variability that exist

when the same character is written by different

persons or same character written at different

times. The feature extraction and stroke

classification address this issue to identify an


80

ISSN:2249-5789

appropriate representation and method of

classification.

2) Which dataset can be used for handwritten

recognition?

The standard database for Indian script is neither

available freely nor commercially, hence we have

collected the samples Devnagari handwritten

characters data from different professional

belonging to schools, colleges, nor commercial sectors are collected and created the data set.

Generation of database from the scanned

datasheets.

3) Which techniques are used to extract the features

for recognition of characters?

Feature extraction is extracting information from

raw data which is most relevant for classification

purpose and that minimizes the variations within a

class and maximizes the variations between

classes. Different feature extraction methods are

designed for different representations of the

characters, such as solid binary characters,

character contours, skeletons or gray level sub

images of each individual character. In above

discussion we have studied that there are so many

features extraction techniques. In that we have

used chain code histogram to character contour and gradient features for recognition of Marathi

characters.

4) Which classifiers are used?

SVM is better than other classifier because of its

complexity of training, flexibility, classification

accuracy and complexity which is given bellow.

8.2 Cloud Based Recognition System

The architecture of the proposed Cloud-

based handwriting recognition system is shown in

Figure 8. The client terminal is Computer device or

laptop with different operating systems, such as

Windows XP, Windows 7 Embedded Linux so on.

There is a scanned image of the handwritten

character through which we take the strokes of input

character. The server is virtual Cloud combining

many physical server machines and personal

computers. When finishing writing, strokes are sent

to the server by socket protocol through WiFi,

GPRS, EDGE or the 3G networks. The handwriting recognition are run on the Cloud server Relative to

the bandwidths of the networks, the data of the

strokes and candidates is small, so it costs very little

transmission time and users have no sense about the

delay. It feels like that the recognition is done locally

In theory, the memory and computing capacity of

Cloud server are infinite. So, we can implement any

state-of-the art recognition algorithm to get very high

accuracy in real time[2].

Figure 8. Architecture of cloud based system

9. Discussion on Review

9.1 Preprocessing

Pre-processing aims to produce data that are

easy for the character recognition systems to operate

accurately. The main objectives of pre-processing

are:

1) Binarization: If the illumination over the

document is not uniform, for instance in the case of

scanned book pages or camera-captured documents,

global binarization methods tend to produce

marginal noise along the page borders. To overcome

these complexities, local thresholding

techniqueshave been proposed for document

binarization[21].

2) Normalization: Normalization is used to reduce the variation of

shapes of the character. Size normalization depends

on how user moves the pen on writing pad. Centering

is required when pen is moved along the border of

writing pad. Normalization is for reducing variability

in character size and pen velocity

3) Smoothing: Smoothing operations are used to blur

the image and reduce the noise. Smoothing and noise

removal can be done by filtering.

4) Interpolation: Input handwritten characters could

have missing points when handwriting speed is fast.

It is used to find missing points in characters.

Bresenham’s line algorithms can be used to estimate

the missing intermediate points.

5) Slant Correction: Slant correction and normalizing

slant is required to correct the shape of input

handwritten character which is bending in left or

right directions. For this slant correction DESLANT algorithm is used. Slant correction method uses the

vertical projection histogram. The idea is that the

histogram of a word that is written straight up will

have larger and more distinct peaks. Therefore, look


81

ISSN:2249-5789

at the histogram of the word at different shear angles

and take the one with the highest peaks. For angles

between −45 and 45 degrees, which is the most

common range of slant angles in normal writing[22].

9.2 Segmentation

In complicated case of handwritten text, the

segmentation problem becomes difficult when letters

tend to be connected to each other, overlapped or

distorted. Segmentation is break the single text line,

single word and single character from the input

document. For isolated characters or numerals,

segmentation task is not that difficult. However, for

joint and complex strings more advanced techniques

required to be employed. Average Longest path algorithm is used in different handwriting styles

large variation of neighboring characters within

words are usually connected for that time need to

segment the word into individual character for

accurate character recognition[23].

9.3 Feature Extraction Techniques

There are various feature extraction

techniques but freeman chain code and gradient

feature extraction techniques are better. The freeman

chain code approach is robustness to small variation

and easy to implement and which provides good

recognition rate. The gradient feature extraction

technique can be easily used to gray scale images

and are robust against image noise and edge

direction variation.

9.4 Classifier

Neural classifiers, K Nearest Neighbor and

SVMs show different properties in the following

respects[24][25].

Complexity of training: The parameters of neural classifiers are generally adjusted by gradient descent.

By feeding the training samples a fixed number of

sweeps, the training time is linear with the number of

samples. K-NN classifiers are generally adjusted by

distance measure. By feeding the training samples a

fixed number of sweeps, the training time is linear

with the number of samples. SVMs are trained by

quadratic programming (QP), and the training time is

generally proportional to the square of number of

samples. Some fast SVM training algorithms with

nearly linear complexity are available, however.

Flexibility of training: The parameters of neural

classifiers can be adjusted in string-level or layout-

level training by gradient descent with the aim of

optimizing the global performance. In this case, the

neural classifier is embedded in the string or layout

recognizer for character recognition. The parameters of K-NN classifiers can be adjusted in Feature

weighting improve classification accuracy for global

performance and also easy to add a new class to an

existing classifier. On the other hand, SVMs can only

be trained at the level of holistic patterns.

Classification accuracy: SVMs have been

demonstrated superior classification accuracies to

neural classifiers in many experiments. The

parameters of K-NN classifiers can be adjusted in

Feature weighting improve classification accuracy for

global performance and also easy to add a new class to an existing classifier.

Storage and execution complexity: SVM learning

by QP often results in a large number of SVs, which

should be stored and computed in classification.

Neural classifiers have much less parameters, and the

number of parameters is easy to control. In a word,

neural classifiers consume less storage and

computation than SVMs. K-NN classifiers have much

less parameters, and are easy to control. In a word, K-

NN classifiers consume less storage and computation

than SVMs.

In above discussion SVM is better than

KNN and ANN classifier because of its complexity

of training, flexibility, classification accuracy and

complexity. The SVM is a machine learning

algorithm which

Solves classification problems

Uses a flexible representation of the class

boundaries

Implements automatic complexity control to reduce

over fitting

Has a single global minimum which can be found

in polynomial time.

SVM is popular because

It can be easy to use

It often has good generalization performance

The same algorithm solves a variety of problems

with little tuning.

Conclusion The idea of regarding handwritten character

recognition as a service based on Cloud computing, it

can be more convenient. These clouds can evolve as

new data is received by the server, improving

recognition. These clouds also provide a simple but

effective method for handwriting recognition. Among

all the methods it is found that chain code histogram

of a character contour and gradient feature extraction

techniques are the best performing with SVM

classifier.

References [1] Vadim Mazalov and Stephen M. Watt,“ Writing on

Clouds”

[2]Yan Gao, Lanwen Jin+, Cong He, Guibin Zhou,”

Handwriting Character Recognition as a Service：A


82

ISSN:2249-5789

New Handwriting Recognition System Based on Cloud

Computing”, ICDAR '11 Proceedings of the 2011 International Conference on Document Analysis and

Recognition, pp. 885-889, 2011 IEEE.

[3]K. Sethi and B. Chatterjee, “Machine Recognition

constrained Hand printed Devanagari”, PatternRecognition, Vol. 9, pp. 9-75, 1977.

[4]U. Pal, B.B. Chaudhury, “Indian Script Character

Recognition: A Survey”, Pattern Recognition, Elsevier,

pp. 1887-1899, 2004. [5]Bikash Shaw, Swapan Kr. Parui, Malayappan Shridhar,“

Offline Handwritten DevanagariWord Recognition:A

Segmentation Based Approach,” 2008 IEEE

[6] Arora, D. Bhattacharjee, M. Nasipuri, D.K. Basu, and M. Kundu, “Combining Multiple Feature Extraction

Techniques for Handwritten Devanagari Character

Recognition IEEE Region 10 Colloquium and Third

Intl. Conf. Industrial & Information Systems, Kharagpur (India), pp. 978-1-4244 2806 2008.

[7] Prachi Mukherji, Priti Rege, “Shape Feature and

FuzzyLogic Based Offline evanagari Handwritten

Optical Character Recognition”, Journal of Pattern Recognition Research 4,pp. 52-68, 2009.

[8] U. Pal, Wakabayashi, Kimura, “Comparative Study of

Devanagari Handwritten Character Recognition using

Different Feature and Classifiers”, 10th International Conference on Document Analysis and Recognition,

pp. 1111-1115, 2009.

[9]Mallikarjun Hangarge, B.V.Dhandra, “Offline

Handwritten Script Identification in Document Images”,

International Journal of Computer Applications (0975

– 8887), July 2010

[10]K. Roy, S. Kundu Das, Sk Md Obaidullah ,“Script Identification from Handwritten Document”, Third

National Conference on Computer Vision, Pattern

Recognition, Image Processing and Graphics, pp.8 -11, 2011 IEEE

[11]Ved Prakash Agnihotri Offline Handwritten

Devanagari Script Recognition”, I.J. Information

Technology and Computer Science, pp. 37-42July 2012 in MECS

[12]Pankaj Kumawat , Asha Khatri , Baluram Nagaria,“

Comparative analysis of offline Handwriting

Recognition Using Invariant Moments with HMM and combined SVM-HMM classifier”, International

Conference on Communication Systems and Network

Technologies, 2013 IEEE

[13]Ms. Vaishali R. Wadhe, Dr. Vinayak A. Bharadi,“ Review: Services and Applications of Cloud

Computing”, International Journal of Scientific &

Engineering Research, pp.979-984 12, December-

2013. [14] Gaurav Y. Tawde , Mrs. Jayashree M. Kundargi, “ An

Overview of Feature Extraction Techniques in OCR

for Indian Scripts Focused on Offline Handwriting”,

International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622

www.ijera.com Vol. 3, Issue 1, January -February

2013, pp.919-926.

[15] Prof. M.S.Kumbhar, Y.Y.Chandrachud, “ Handwritten marathi character Recognition using neural

network”,International Journal of Emerging

Technology and Advanced Engineering.

[16] S. Arora , D. Bhattacharjee, M. Nasipuri2 , D.K. Basu,

M.Kundu, “Application of Statistical Features in Handwritten Devnagari Character Recognition”.

[17] Kai Ding, Zhibin Liu, Lianmen Jin, Xinghua Zhu, “A

Comparative Study of Gabor Feature and Gradient

Feature For handwritten chinese character recognition

[18] Sneha U.Bohra, Satish J.Alaspurkar,“Handwritten

Character Recognition for major Indian Scripts:A

Survey”, Sneha U.Bohra et al./ International Journal of Computer Science & Engineering Technology

(IJCSET), ISSN : 2229-3345,pp.354-364, Apr 2013.

[19] Brad Jannsen, “Support Vector Machines for Binary

Classification and its Applications”. [20] Sanghamitra Mohanty, Himadri Nandini Das Bebartta,

“Performance Comparison of SVM and K-NN for

Oriya Character Recognition.

[21] Øivind Due Trier and Torfinn Taxt, “Evaluation of binarization methods valuation of Binarization

Methods for Document Images”.

[22] Pooja Tewari, “Use of Dominant Point detection

Feature for Recognition of Online Handwritten Devanagari Script”, June 2012

[23] Dhaval Salvi, Jun Zhou, Jarrell Waggoner, and Song

Wang, “Handwritten Text Segmentation using

Average Longest Path Algorithm”, 978-1-4673-5052 IEEE pp. 505 – 512, 2012.

[24] Sandhya Arora, Debotosh Bhattacharjee, Mita

Nasipuri, L. Malik , M. Kundu and D. K. Basu, “

Performance Comparison of SVM and ANN for Handwritten Devanagari Character Recognition”,

IJCSI International Journal of Computer Science

Issues, Vol. 7, Issue 3, May 2010.

[25] Sanghamitra Mohanty, Himadri Nandini Das Bebartta, “Performance Comparison of SVM and K-NN for

Oriya Character Recognition”, IJACSA)

International Journal of Advanced Computer Science

and Applications, Special Issue on Image Processing and Analysis, pp.112-116.


83

ISSN:2249-5789

offline marathi handwritten character recognition … marathi handwritten character recognition...

Documents