Ancient Tamil Script Language Recognition In Noisy And Degraded Stone Inscription Image

Abstract

Ancient Tamil character recognition is one of the most challenging tasks in the pattern recognition system. It continues to be an active area of research towards exploring the newer techniques that would improve recognition accuracy. The methodologies proposed towards step in ancient Tamil character recognition system is pre- processing followed by segmentation and feature extraction, feature selection techniques are applied to enhance classification. newline newlineIn pre-processing the image file is checked for skewing, if the image is skewed it is corrected by a simple rotation technique in the appropriate direction, then the image is passed through salt and pepper noise and elimination using median filtering technique which threshold the process of binarization. The accuracy of recognition will be increased with the use of very good segmentation which decomposes the scanned text into a paragraph using spatial space detection technique and then paragraphs into lines using vertical histogram and lines into words using horizontal histogram and words into character image using horizontal histograms. Each image is subjected to feature extraction such as character height, character width, number of horizontal lines, number of vertical lines, horizontally oriented curves, vertical oriented curves, number of circles, number of slope lines, image centroid, and special newline newline newline newlinedots using Shape and Hough transform methods of feature extraction are used in this work. The proposed work is to concatenate features that help to obtain a better feature subset. This is a very important stage for the recognition system. It is a challenging task for the computation of good features. newline newlineThe feature selection will help in the improvement and efficiency of the classifiers in this work, optimized feature selection FF, GSO, and Hybrid GSO-FF are presented. The FF and GSO are existing heuristic algorithms. We adapt the existing for feature selection and a novel fitness function is proposed in serial. newline newlineThe classifiers such as J48, K Nearest Neighbor