Analysis and Representation of Character Images for Extracting Shape based Features Towards Building an OCR for Bangla Script

Abstract

In this paper we present the analysis of different representation techniques of the character images for extracting shape based features using modified signature method. An adaptive normalization is proposed that retains the aspect ratio of the character which is subjected to the fact that preservation of aspect ratio enhances the recognition process. Signature is generated from direction angle of the pixels where the pixels were taken from boundary, difference and skeletonized image of the character. Correlation among the extracted features for 13 different fonts of all 50 characters is analyzed. We measured the recognition performance of different representation approaches.