V. N. Manjunath Aradhya

A novel approach for biclustering gene expression data using modular singular value decomposition

Clustering is a method of unsupervised learning, and a common technique for statistical data analysis used in many fields, including machine learning, data mining, pattern recognition, image analysis and bioinformatics. Recently, biclustering (or co-clustering), performing simultaneous clustering on the row and column dimensions of the data matrix, has been shown to be remarkably effective in a variety of applications. In this paper we propose a novel approach to biclustering gene expression data based on Modular Singular Value Decomposition (Mod-SVD). Instead of applying SVD directly on on data matrix, the proposed approach computes SVD on modular fashion. Experiments conducted on synthetic and real dataset demonstrated the effectiveness of the algorithm in gene expression data.

An Application of K-Means Clustering for Improving Video Text Detection

In the present work, we explore an extensive applications of Gabor filter and K-means clustering algorithm in detection of text in an unconstrained complex background and regular images. The system is a comprehensive of four stages: In the first stage, combination of wavelet transforms and Gabor filter is applied to extract sharpened edges and textural features of a given input image. In the second stage, the resultant Gabor output image is grouped into three clusters to classify the background, foreground and the true text pixels using K-means clustering algorithm. In the third stage of the system, morphological operations are performed to obtain connected components, then after a concept of linked list approach is in turn used to build a true text line sequence. In the final stage, wavelet entropy is imposed on an each connected component sequence, in order to determine the true text region of an input image. Experiments are conducted on 101 video images and on standard ICDAR 2003 database. The proposed method is evaluated by testing the 101 video images as well with the ICDAR 2003 database. Experimental results show that the proposed method is able to detect a text of different size, complex background and contrast. Withal, the system performance outreaches the existing method in terms of detection accuracy.

A particle swarm optimization method for tuning the parameters of multiscale retinex based color image enhancement

In this paper, a Particle Swarm Optimization (PSO) method for tuning the parameters of multiscale retinex based color image enhancement is presented. The image enhancement using multiscale retinex scheme heavily depends on parameters such as Gaussian surround space constants, number of scales, gain and offset etc. Due to hard selection of these parameters, PSO has been used in order to investigate the optimal parameters for the best image enhancement. The PSO method of parameter tuning adopted for multiscale retinex with modified color restoration (MSRMCR) algorithm achieves very good quality of reconstructed images, far better than that possible with the other existing methods. The presented algorithm is compared with other promising enhancement schemes such as histogram equalization, NASAs multiscale retinex with color restoration (MSRCR), Improved MSRCR (IMSRCR), and Photoflair software. The quality of the enhanced image is validated iteratively using an efficient objective criterion which is based on entropy and edge information of an image. Finally, the quality of the reconstructed images obtained by the proposed method is evaluated using Wavelet Energy (WE) metric. The experimental results presented shows that color image enhanced by the proposed algorithm are clearer, vivid and efficient.

Text line segmentation of unconstrained handwritten Kannada script

Separating text lines in handwritten documents remains a challenge because the text lines are often varying skewed and curved. In this paper, we propose a novel method for text line segmentation of unconstrained handwritten Kannada script. The proposed method consists of two phases. In the first phase, mathematical morphology technique is used to bridge the gap between character components. In the Second phase, component extension technique is used for text line extract. We experimentally evaluated our proposed method on document containing handwritten Kannada script and showed encourage results.

An impact of ridgelet transform in handwritten recognition: A study on very large dataset of Kannada script

Handwritten character recognition is a difficult problem due to the great variations on writing styles, different size and orientation angle of the characters. In this paper, we propose an unconstrained handwritten Kannada character recognition based on the ridgelet transforms. Ridglets are a powerful instrument in catching and representing mono-dimensional singularities in bi dimensional space [7]. Ridgelet transforms is used to extracts low pass energy of character image and is then fed to PCA for feature extraction. We conducted experiment on very large database of handwritten Kannada character. The size of the class was 200 and encouraging results are obtained.

Handwritten character segmentation for Kannada scripts

Character segmentation has become a crucial task for character recognition in many OCR systems. It is an important step because incorrectly segmented characters are unlikely to be recognized correctly. For segmenting a cursive scripts leads more challenging because of presence of more touching characters. Kannada is the one of the popular language in south India and also some of the letters in Kannada language are cursive in nature. In this paper, a new character segmentation algorithm for unconstrained handwritten Kannada scripts is presented. The proposed method is based on thinning, branch point and mixture models. The expectation-maximization (EM) algorithm is used to learn the mixture of Gaussians. We have used a cluster mean points to estimate the direction and branch point as reference points for segmenting characters. We have experimentally evaluated our proposed method on Kannada words and it has shown encouraging result.

The study of different similarity measure techniques in recognition of handwritten characters

In this paper, we compare the affect of four different similarity measure techniques namely Euclidean distance, Modified squared euclidean distance, Correlation distance and Angle distance for an unconstrained handwritten character recognition. The strength of these similarity measures are estimated between feature vectors with respect to the recognition performance of the Gabor-PCA method. Gabor filter is used to extract spatially localized features of character image. The dimensions of such Gabor feature vector is prohibitively high & in order to compress Gabor features we used PCA method. The experiments were performed using the database containing 22,600 samples of Kannada and English. From the analysis the better recognition accuracy were achieved using angle distance measure.

An Improved Handwritten Text Line Segmentation Technique

Document image segmentation to textlines is a decisive stage towards unconstrained handwritten document recognition. In this paper, an improvised scheme for handwritten text line segmentation is proposed. The proposed method is an improvised method to Alaei et al., 2010 [1] by applying block separation and edge detection process. The proposed method was tested on variety of handwritten documents pertaining to English, Persian, and Kannada and the results was remarkable.

Document skew estimation: an approach based on wavelets

Document skew estimation refers to the process of finding the angle of inclination made by the document with respect to horizontal axis. The skew introduced during the scanning process like this is inevitable, even slightest degree of skew will always be there irrespective of how the document is fed to the scanner: either manually or automatically. Hence, deskewing of the document is vital for achieving efficient results in downstream document analysis system (DAS) such as page layout analysis, optical character recognition (OCR), document retrieval etc. Although enormous amount of research has been conducted for document skew estimation, development of a solitary skew estimation approach that can handle all kinds of real time variation in documents is still an elusive goal for the research community. In this paper, we present a novel scheme for estimating document skew based on Wavelets. In the first stage, document images are moldered by the wavelet transform and efficient hough transform is used for estimating the skew of a document. Experimental results show that the method performs well on document images of complex layouts and to different scripts.

Skew Estimation for Unconstrained Handwritten Documents

Document skew estimation is one of the most important and challenging phase in OCR system. Skew estimation of handwritten documents is still remains challenging in the field of document image analysis due to a non-uniform text line. Hence, in this paper, we present a novel scheme for handwritten documents. The proposed method is based on mixture models. The expectation-maximization (EM) algorithm is used to learn the mixture of Gaussians. Subsequently the cluster means obtained from the individual words is used estimate the skew angle. Experiments on different handwritten documents and documents corrupted by noise shows the effectiveness of the proposed method.