DIMENSION REDUCTION FOR SCRIPT CLASSIFICATION- PRINTED INDIAN DOCUMENTS

International Journal of Advanced Information Technology (IJAIT) Vol. 7, No.1/2/3, June 2017 DOI : 10.5121/ijait.2017.7301 1 D IMENSION R EDUCTION FOR S CRIPT C LASSIFICATION - P RINTED I NDIAN D OCUMENTS Hamsaveni L 1 , Pradeep C 2 and Chethan H K 3 1 Department of Studies in Computer Science Manasagangotri, University of Mysore, Mysore 2 Department of Computer Science and Engineering Rajarajeswari College of Engineering, Bangalore 3 Department of Computer Science and Engineering Maharaja Institute of Technology, Mysore A BSTRACT Automatic identification of a script in a given document image facilitates many important applications such as automatic archiving of multilingual documents, searching online archives of document images and for the selection of script specific OCR in a multilingual environment. This paper provides a comparison study of three dimension reduction techniques, namely partial least squares (PLS), sliced inverse regression (SIR) and principal component analysis (PCA), and evaluates the relative performance of classification procedures incorporating those methods. For given script we extracted different features like Gray Level Co-occurrence Method (GLCM) and Scale invariant feature transform (SIFT) features. The features are extracted globally from a given text block which does not require any complex and reliable segmentation of the document image into lines and characters. Extracted features are reduced using various dimension reduction techniques. The reduced features are fed into Nearest Neighbor classifier. Thus the proposed scheme is efficient and can be used for many practical applications which require processing large volumes of data. The scheme has been tested on 10 Indian scripts and found to be robust in the process of scanning and relatively insensitive to change in font size. This proposed system achieves good classification accuracy on a large testing data set. K EYWORDS SIFT, GLCM, PLS, SIR, PCA, Nearest Neighbour 1. I NTRODUCTION Document image analysis has been an active research area from a few decades, and that facilitates the establishment of paperless offices across the world. The process of converting textual symbols present on printed and/ or handwritten paper to a machine understandable format is known as optical character recognition (OCR) which is the core of the field of document image analysis. The OCR technology for Indian documents is in emerging stage and most of these Indian OCR systems can read the documents written in only a single script. As per the trilingual formula of Indian constitution [1], every state Government has to produce an official document containing a national language (Hindi), official language (English) and state language (or regional language). According to the three-language policy adopted by most of the Indian states, the documents produced in an Indian state Karnataka, are composed of texts in the regional language-Kannada, the National language-Hindi and the world wide commonly used language-English. In addition, majority of the documents found in most of the private and Government sectors of Indian states, are tri-lingual type (a document having text in three languages). So, there is a growing demand to automatically process these tri-lingual documents in every state in India, including Karnataka. 2 The monolingual OCR systems will not process such multi-script documents without human involvement for delineating different script zones of multi-lingual pages before activating the script specific OCR engine. The need for such manual involvement can result in greater expense and crucially delays the overall image-to-text conversion. Thus, an automatic forwarding is required for the incoming document images to handover this to the particular OCR engine depending on the knowledge of the intrinsic scripts. In view of this, identification of script and/ or language is one of the elementary tasks for multi-script document processing. A script recognizer, therefore, simplifies the task of OCR by enhancing the accuracy of recognition and reducing the computational complexity. 2. P REVIOUS W ORK Existing works on automatic script identification are classified into either local approach or global approach. Local approaches extract the features from a list of connected components like line, word and character in the document images and hence they are well suited to the documents where the script type differs at line or word level. In contrast, global approaches employ analysis of regions comprising of at least two lines and hence do not require fine segmentation. Global approaches are applicable to those documents where the whole document or paragraph or a set of text lines is in one script only. The script identification task is simplified and performed faster with the global rather than the local approach. A sample work has been reported in literature on both Indian and non-Indian scripts using local and global approaches. 2.1 Local approaches on Indian scripts Pal and Choudhuri [2] have proposed an automatic technique of separating the text lines from 12 Indian scripts (English, Hindi, Bangla, Gujarati, Tamil, Kashmiri, Malayalam, Oriya, Punjabi, Telugu and Urdu) using ten triplets formed by grouping English and Devanagari with any one of the other scripts. This method works only when the triplet type of the document is known. Script identification technique explored by Pal [3] uses a binary tree classifier for 12 Indian scripts using a large set of features. B Patil and Subbareddy [4] have proposed a neural network based system for script identification of Kannada, Hindi and English languages. Dhandra et al., [5] have exploited the use of discriminating features (aspect ratio, strokes, eccentricity, etc,) as a tool for determining the script at word level in a bi-lingual document containing Kannada, Tamil and Devnagari containing English numerals. A method to automatically separate text lines of Roman, Devanagari and Telugu scripts has been proposed by Pal et al., [6]. In Lijun et al, [7] have developed a method for Bangla and English script identification based on the analysis of connected component profiles. Vipin [8] have presented an approach to automatically identify Kannada, Hindi and English languages using a set of features viz., cavity analysis, end point analysis, corner point analysis, line based analysis and Kannada base character analysis. Wordwise script identification systems for Indian scripts has been discussed in [24]. 2.2 Global approaches on Indian scripts Adequate amount of work has been reported in literature using global approaches. S Chaudhury et al., [9] has proposed a method for identification of Indian languages by combining Gabor filter based technique and direction distance histogram classifier considering Hindi, English, Malayalam, Bengali, Telugu and Urdu. G D Joshi et al., [10] have presented a script identification technique for 10 Indian scripts using a set of features extracted from logGabor filters. Dhanya et al., [11] have used Linear Support Vector Machine (LSVM), K-Nearest Neighbour (K-NN) and Neural Network (NN) classifiers on Gabor-based and zoning features to classify Tamil and 3 English scripts. Hiremath [12] have proposed a novel approach for script identification of South Indian scripts using wavelet based co-occurrence histogram features. Ramachandra and Biswas [13] have proposed a method based on rotation invariant texture features using multi channel Gabor filter for identifying seven Indian languages namely Bengali, Kannada, Malayalam, Oriya, Telugu and Marathi. S R Kunte and S Samuel [14] have suggested a neural approach in on-line script recognition for Telugu language employing wavelet features. Nagabhushan et al., [15] have presented an intelligent pin code script identification methodology based on texture analysis using modified invariant moments. Peeta et al., [16] have presented a technique using Gabor filters for script identification of Indian bilingual documents. 2.3 Local and global approaches on non-Indian scripts Sufficient amount of work has also been carried out on non-Indian languages. Spitz [17] has proposed a system, which relies on specific, well defined pixel structures for script identification. Such features include locations and numbers of upward concavities in the script image, optical density of connected components, the frequency and combination of relative character heights. This approach has been shown to be successful in distinguishing between Asian languages (Japanese, Chinese, and Korean) against European languages (English, French, German, and Russian). Wood et al., [18] have proposed projection profile method to determine Roman, Russian, Arabic, Korean and Chinese characters. Hochberg et al., [19] have presented a method for automatically identifying script from a binary document image using cluster-based text symbol templates. In Ding et al., [20], a method that uses a combined analysis of several discriminating statistical features to classify Oriental and European scripts is presented. Tan et al., [21] has proposed a rotation invariant texture feature extraction method for automatic script and language identification from document images using multiple channel (Gabor) filters and Gray level co-occurrence matrices for seven languages: Chinese, English, Greek, Koreans, Malayalam, Persian and Russian. A Busch et al., [22] has presented the use of texture features (gray level cooccurrence matrix and Gabor energy features) for determining the script of a document image. B.Kumar et al. [23] have used topological, structural features with rule based classifier for line based multi-script identification. It can be seen from the references cited above that sample amount of work has been done in the area of document script/language identification. Even though some considerable amount of work has been carried out on Indian script identification, hardly few attempts focus on the all the languages. So, an intensive work needs to be done in this field as the demand is increasing. Also the existing methods have to be improved to reach a stage of satisfactory practical application. It is in this direction the research work proposes a model that automatically identifies the all the languages in given document. We propose a based classification scheme which uses a global approach and demonstrate its ability to classify 10 Indian language scripts. In section (3), we describe the preprocessing scheme. The feature extraction is presented in section 4. The various dimension reduction techniques is discussed in section 5. Results of the scheme tested over a large data set are presented in section 6. 3. P REPROCESSING Our scheme first segments the text area from the document image by removing the upper, lower, left and right blank regions. After this stage, we have an image which has textual and non-textual regions. This is then binarised after removing the graphics and pictures (at present the removal of non-textual information is performed manually, though page segmentation algorithms such as [12] could be readily been employed to perform this automatically). Text blocks of predefined size (100×200 pixels) are next extracted. It should be noted that the text block may contain lines 4 with different font sizes and variable spaces between lines words and characters. Numerals may appear in the text. 4. F EATURE E XTRACTION Feature extraction is a necessary step for any classification task. For image object classification purpose, the use of texture and shape features has proved to be quite effective for many applications. There are many ways for calculating texture feature descriptors. GLCM is one of them. Many descriptors can be obtained from the co-occurrence matrix calculated. The SIFT based descriptors describes a given object with respect to a set of interesting points which are invariant to scale, translation, partial occlusion and clutter. These feature descriptors have been used successfully for object recognition, robotic mapping etc. In our work, for each script, we computed 4 texture features, contrast, homogeneity, correlation and energy. For each object, the SIFT algorithm generates a feature vector of 128 elements. So each image object is now represented by a feature vector of 132 elements. 4.1 GLCM Based Texture Feature Descriptors Texture features based on spatial co-occurrence of pixel values are probably the most widely used texture feature descriptors having been used in several application domains like analysis of remotely sensed images, image segmentation etc. Cooccurrence texture features are extracted from an image into two steps. First, pair wise spatial co-occurrence of pixels separated by a given angular value are computed and stored in a grey level co-occurrence matrix. Second, the GLCM is used to compute a set of scalar quantities that characterizes the different aspects of the underlying texture. We have worked with four GLCM based descriptors, namely, Contrast, Correlation, Homogeneity and Energy [26]. 4.2 SIFT Feature Descriptors In computer vision, SIFT is used to detect and describe local features in an image. SIFT features are used for reliable matching between different views of the same object. The extracted features are invariant to scale, orientation and are partially invariant to illumination changes. The SIFT feature extraction is a four step process. In the first step, locations of the potential interest points are computed in the image by finding the extremas in a set of Difference of Gaussian (DOG) filters applied to the actual image at different scale-space. Then those interest points which are located at the areas of low brightness and along the edges are discarded. After that an orientation is assigned to the remaining points based on local image gradients. Finally local image features based on image gradient is calculated at the neighboring regions of each of the key points. Every feature is defined in the 4 x 4 neighborhoods of the key points and is a vector of 128 elements [27]. 5. D IMENSION R EDUCTION The extracted features are reduced using various dimension reduction techniques. One way to achieve dimension reduction is to transform the large number of original variables (genes) to a new set of variables (gene components), which are uncorrelated and ordered so that the first few account for most of the variation in the data. The K new variables (gene components) can then replace the initial p variables (genes), thereby reducing the data from the high p-dimension to a lower K-dimension. PCA, PLS and SIR are three of such methods for dimension reduction. To 5 describe them, let X be the n × p matrix of n samples and p features, y be the n × 1 vector of response values, and SX be the p × p covariance matrix. 5.1 Principal Component Analysis PCA is a well-known method of dimension reduction (Jolliffe, [30]). The basic idea of PCA is to reduce the dimensionality of a data set, while retaining as much as possible the variation present in the original predictor variables. This is achieved by transforming the p original variables X = [x1, x2, …, xp] to a new set of K predictor variables, T = [t1, t2, …, tK], which are linear combinations of the original variables. In mathematical terms, PCA sequentially maximizes the variance of a linear combination of the original predictor variables, )(maxarg 1' uuuk XVaru = = (1) subject to the constraint ui 'SX uj = 0, for all 1 ≤ i < j. The orthogonal constraint ensures that the linear combinations are uncorrelated, i.e.Cov(Xui ,Xuj ) = 0, i ≠ j. These linear combinations ij Xut = (2) are known as the principal components (PCs) (Massey, [31]). Geometrically, these linear combinations represent the selection of a new coordinate system obtained by rotating the original system. The new axes represent the directions with maximum variability and are ordered in terms of the amount of variation of the original data they account for. The first PC accounts for as much of the variability as possible, and each succeeding component accounts for as much of the remaining variability as possible. Computation of the principal components reduces to the solution of an eigenvalue-eigenvector problem. The projection vectors (or called the weighting vectors) u can be obtained by eigenvalue decomposition on the covariance matrix SX, S X ui = λi ui (3) where λi is the i-th eigenvalue in the descending order for i=1,…,K, and ui is the corresponding eigenvector. The eigenvalue λi measures the variance of the i-th PC and the eigenvector ui provides the weights (loadings) for the linear transformation (projection). The maximum number of components K is determined by the number of nonzero eigenvalues, which is the rank of S X, and K ≤ min(n,p). The computational cost of PCA, determined by the number of original predictor variables p and the number of samples n, is in the order of min(np2 + p3 , pn2 + n3). In other words, the cost is O(pn2 + n3) when p > n. 5.2 Partial Least Squares The objective of constructing components in PLS is to maximize the covariance between the response variable y and the original predictor variables X, w K = arg max Cov(Xw, y) (4) w'w=1 subject to the constraint wi 'S X w j = 0 , for all 1 ≤ i < j. The central task of PLS is to obtain the vectors of optimal weights wi (i=1,…,K) to form a small number of components that best predict the response variable y. Note that PLS is a “supervised” method because it uses information on both X and y in constructing the components, while PCA is an “unsupervised” method that utilizes the X data only. 6 To derive the components, [t1, t2, …, tK], PLS decomposes X and y to produce a bilinear representation of the data (Martens and Naes, 1989): X = t1w'1 +t2 w'2 +... + t K w'K +E (5) and y = t1q1 + t2 q2 +... + t K qK + F (6) where w’s are vectors of weights for constructing the PLS components t=Xw, q’s are scalars, and E and F are the residuals. The idea of PLS is to estimate w and q by regression. Specifically, PLS fits a sequence of bilinear models by least squares, thus given the name partial least squares (Wold, [32],[33],[34]). At each step i (i=1,…,K), the vector wi is estimated in such a way that the PLS component, ti, has maximal sample covariance with the response variable y subject to being uncorrelated with all previously constructed components. The first PLS component t1 is obtained based on the covariance between X and y. Each subsequent component ti (i=2,…,K), is computed using the residuals of X and y from the previous step, which account for the variations left by the previous components. As a result, the PLS components are uncorrelated and ordered (Garthwaite, [35]; Helland, [36], [37]). The maximum number of components, K, is less than or equal to the smaller dimension of X, i.e. K ≤ min(n,p). The first few PLS components account for most of the covariation between the original predictors and the response variable and thus are usually retained as the new predictors. The computation of PLS is simple and a number of algorithms are available (Martens and Naes, [38]). In this study, we used a standard PLS algorithm (Denham, [39]). Like PCA, PLS reduces the complexity of microarray data analysis by constructing a small number of gene components, which can be used to replace the large number of original gene expression measures. Moreover, obtained by maximizing the covariance between the components and the response variable, the PLS components are generally more predictive of the response variable than the principal components. The number of components, K, to be used in the class prediction model is considered to be a meta parameter and must be estimated in the application, which we will discuss later. PLS is computationally very efficient with cost only at O(np), i.e. the number of calculations required by PLS is a linear function of n and p. Thus it is much faster than the other two methods (PCA and SIR). 5.3 Sliced Inverse Regression SIR, one of the sufficient dimension reduction methods (Li, [40]; Duan and Li,[41];Cook [42]), is a supervised approach, which utilizes response information in achieving dimension reduction. The idea of SIR is simple. Conventional regression models deal with the forward regression function, E(y|X), which is a p-dimensional problem and difficult to estimate when p is large. SIR is based on the inverse regression function, η(y) = E(X | y) (7) which consists of p one-dimensional regressions and is easier to deal with. The SIR directions v can be obtained as the solution of the following optimization problem, 7 vK = arg max v' Cov(E(X | y))v (8) v' SX v v'v=1 subject to the constraint vi 'S X v j = 0 , for all 1 ≤ i < j. Algebraically, the SIR components ti=Xvi ( i=1,…,K) are linear combinations of the p original predictor variables defined by the weighting vectors vi. Geometrically, SIR projects the data from the high pdimensional space to a much lower K-dimensional space spanned by the projection vectors v. The projection vectors v are derived in such a way that the first a few represent directions with maximum variability between the response variable and the SIR components. Computation of vi is straightforward. Let Sη = Cov(E(X | y)) be the covariance matrix of the inverse regression function defined in (7) and recall that SX is the variance-covariance matrix of X. The vectors vi (i=1,…,K) can be obtained by spectral decomposition of Sη with respect to S X , Sη vi = λiS X vi (9) where λi is the i-th eigenvalue in descending order for i=1,…,K; vi is the corresponding eigenvector, and vi 'S X v j =1. SIR is implemented by appropriate discretization of the response. Let T(y) be a discretization of the range of y. SIR computes Cov(E(X|T(y))), the covariance matrix for the slice means of X, which can be thought of as the between covariance for the subpopulations of X defined by T(y). Usually, if the response is continuous, one divides its range into H slices. If the response is categorical, one simply considers its categories. In class prediction problems, the number of classes G is a natural choice for H, i.e. H=G. The maximum number of SIR components is H minus one, i.e. K ≤ min(H-1,n,p). As discussed before, K is considered to be a meta-parameter and may be estimated by cross-validation. The cost of computing SIR directions using the standard algorithm is O(np2 + p3), which is quite expensive comparing to the cost of PLS. We used a standard SIR algorithm (Härdle et al., [43]) in this study. 6. E XPERIMENTS AND R ESULTS 6.1 Data Collection At present, in India, standard databases of Indian scripts are unavailable. Hence, data for training and testing the classification scheme was collected from different sources. These sources include the regional newspapers available online [24] and scanned document images in a digital library [25]. 6.1.1 Indian Language Scripts India has 18 official languages which includes Assamese, Bangla, English, Gujarati, Hindi, Konkanai, Kannada, Kashmiri, Malayalam, Marathi, Nepali, Oriya, Punjabi, Rajasthani, Sanakrit, Tamil, Telugu and Urdu. All the Indian languages do not have the unique scripts. Some of them use the same script. For example, languages such as Hindi, Marathi, Rajasthani, Sanskrit and Nepali are written using the Devanagari script; Assamese and Bangla languages are written using the Bangla script; Urdu and Kashmiri are written using the same script and Telugu and Kannada use the same script. In all, ten different scripts are used to write these 18 languages. These scripts are named as Bangla, Devanagari, Roman (English), Gurumukhi, Gujarati, Malayalam, Oriya, 8 Tamil, Kannada and Urdu. The image blocks of these images are shown in Fig. 1. The dataset consists of 10 classes of various scripts, with 100 images of each. 6.2 Nearest Neighbour (NN) One of the simplest classifiers which we used is the Nearest Neighbor classifier [28][29]. The term of nearest can be taken to mean the smallest Euclidean distances in n-dimensional feature space. This takes a test sample feature in vector form, and finds the Euclidean distance between this and the vector representation of each training example. The training sample closest to the test sample is termed its Nearest Neighbor. Since the trained sample in some sense is the one most similar to our test sample, it makes sense to allocate its class label to the test sample. This exploits the ‘smoothness’ assumption that samples near each other are likely to have the same class. 6.3 Results We have performed experiments with different types of images such as normal, bold, thin, small, big, etc. The experimentation has been conducted under varying training samples from 30 to 70 percent of database. We report the accuracies obtained in all cases. The results obtained for various features like SIFT, GLCM and combination (SIFT + GLCM) are respectively shown in Figure 1, Figure 2 and Figure 3. From figures we can understand that the combination of GLCM and SIFT gives a good classification accuracy of 93. 7. C ONCLUSION In this paper, we have proposed a Nearest Neighbour based script classification method with the use features such as GLCM, and SIFT. Specifically, we compared three dimension reduction methods (PLS, SIR, PCA), examined the relative performance of classification procedures incorporating those methods. We found that PLS and SIR were both effective in dimension reduction and they were more effective than PCA. The PLS and SIR based classification procedures performed consistently better than the PCA based procedure in prediction accuracy. The empirical results are consistent with the analysis of the techniques. PLS and SIR construct new predictors using information on the response variable while PCA does not; thus PLS and SIR components are more likely to be good predictors than those from PCA. Considering predictive accuracy, we conclude that the SIR based procedure has provided the best performance among the three classification procedures. R EFERENCES [1]. U. Pal, S. Sinha and B. B.Chaudhri, (2003) “Multi-Script Line Identification from Indian Documents”, Proceedings of International Conference on Document Analysis and Recognition, pp. 880-884. [2]. Pal U., Chaudhuri B.B., (1999), “Script line separation from Indian multi-script document”, Proc. 5th Int. Conf. on Document Analysis and Recognition (IEEE Comput. Soc. Press), 406–409. [3]. Pal U. and Chaudhuri B.B., (2003), “Script line identification from Multi script documents”, IETE journal Vol. 49, No 1, 3-11. [4]. Basavaraj Patil S. and Subbareddy N.V., (2002), “Neural network based system for script identification in Indian documents”, Sadhana Vol. 27, Part 1, 83–97. [5]. Dhandra B.V., Nagabhushan P., Mallikarjun Hangarge, Ravindra Hegadi, Malemath V.S., (2006), “Script Identification Based on Morphological Reconstruction in Document Images”, The 18th International Conference on Pattern Recognition (ICPR'06), Vol.No. 11-3, 950-953. 9 [6]. Pal U., Chaudhuri B.B., (1999), “Script line separation from Indian multi-script document, Proc. 5th Int. Conf. on Document Analysis and Recognition”, (IEEE Comput. Soc. Press), 406–409. [7]. Lijun Zhou, Yue Lu and Chew Lim Tan, (2006),” Bangla/English Script Identification based on Analysis of Connected component Profiles”, Proc. 7th IAPR workshop on Document Analysis System, New land, 234-254. [8]. Vipin Gupta, G.N. Rathna, K.R. Ramakrishnan, (2006), “ A Novel Approach to Automatic Identification of Kannada, English and Hindi Words from a Trilingual Document”, Int. conf. on Signal and Image Processing, Hubli, pp. 561-566. [9]. Santanu Chaudhury, Gaurav Harit, Shekar Madnani, Shet R.B., (2000),” Identification of scripts of Indian languages by Combining trainable classifiers”, Proc. of ICVGIP, India. [10]. Gopal Datt Joshi, Saurabh Garg, and Jayanthi Sivaswamy, (2006), “Script Identification from Indian Documents”, H. Bunke and A.L. Spitz (Eds.): DAS 2006, LNCS 3872, 255–267. [11]. Dhanya D., Ramakrishnan A.G. and Pati P.B., (2002), “Script identification in printed bilingual documents, Sadhana”, vol. 27, 73-82. [12]. Hiremath P S and S Shivashankar, (2008), “Wavelet Based Co-occurrence Histogram Features for Texture Classification with an Application to Script Identification in a Document Image”, Pattern Recognition Letters 29, pp 1182-1189. [13]. Srinivas Rao Kunte R. and Sudhakar Samuel R.D., (2002), A Neural Approach in On-line Script Recognition for Telugu Language Employing Wavelet Features, National Workshop on Computer Vision, Graphics and Image Processing (WVGIP), 188-191. [14]. Peeta Basa Pati, S. Sabari Raju, Nishikanta Pati and A. G. Ramakrishnan, (2004 ), “Gabor filters for Document analysis in Indian Bilingual Documents”, 0-7803-8243-9/04/ IEEE, IClSlP, pp. 123126. [15]. Spitz A. L., (1994), Script and language determination from document images, Proc. of the 3rd Annual Symposium on Document Analysis and Information Retrieval, Las Vegas, Nevada, 229-235. [16]. Wood S. L.; Yao X.; Krishnamurthy K. and Dang L., (1995): Language identification for printed text independent of segmentation, Proc. Int. Conf. on Image Processing, 428–431, IEEE 08186-73109/95. [17]. Hochberg J., Kerns L., Kelly P. and Thomas T., (1997), Automatic script identification from images using cluster based templates, IEEE Trans. Pattern Anal. Machine Intell. Vol. 19, No. 2, 176–181. [18]. Ding J., Lam L. and Suen C. Y., (1997), Classification of oriental and European Scripts by using Characteristic features, Proc. 4th ICDAR , 1023-1027. [19]. Tan T. N., (1998): Rotation invariant texture features and their use in automatic script identification, IEEE Trans. Pattern Anal. Machine Intell. PAMI, Vol.20, No. 7, 751–756. [20]. Andrew Busch; Wageeh W. Boles and Sridha Sridharan, (2005), Texture for Script Identification, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 27, No. 11, pp. 1720-1732. [21]. B. Kumar, A. Bera and T. Patnaik, (2012), “Line Based Robust Script Identification for Indian Languages”, International Journal of Information and Electronics Engineering, vol. 2, no. 2 ,pp. 189192. [22]. R. Rani, R. Dhir and G. S. Lehal, (2013), “Modified Gabor Feature Extraction Method for Word Level Script IdentificationExperimentation with Gurumukhi and English Scripts”, International Journal of Signal Processing, Image Processing and Pattern Recognition, vol. 6, no. 5, pp. 25-38. [23]. A. K. Jain and Y. Zhong., (1996), Page segmentation using texture analysis. Pattern Recognition 29, 743–770. [24]. http://www.samachar.com/. [25]. Digital Library of India. http://dli.iiit.ac.in/ [26]. R. M. Haralick, K. Shanmugam, and I. Dinstein, (1973), Textural Features of % Image Classification, IEEE Transactions on Systems, Man and Cybernetics, % vol. SMC-3, no. 6. [27]. Lowe, D. G., (2004), “Distinctive Image Features from Scale-Invariant Keypoints”, International Journal of Computer Vision, 60, 2, pp. 91-110. [28]. Hall P, Park BU, Samworth R J . (2008), “Choice of neighbor order in nearest-neighbor classification”. Annals of Statistics36 (5): 2135–2152. [29]. Bremner D, Demaine E, Erickson J, Iacono J, Langerman S, Morin P, Toussaint G. (2005), “Outputsensitive algorithms for computing nearest-neighbor decision boundaries”. Discrete and Computational Geometry33 (4): 593–604. [30]. Jolliffe, I.T. (1986), “Principal Component Analysis”. Springer, New York.