Bengali Printed Character Recognition Using A Feature Based Chain Code Method

Abstract

Bengali, one of the official languages of the Indian subcontinent, is composed of 50 alphabets, of which 11 are vowels and 39 consonants. In addition, Bengali words are formed from compound characters and modifiers. Compound characters are formed by combining parts of single characters and modifiers are parts of vowels and consonants which make sense only when adjacent to or attached with a letter. In this paper, features of Bengali characters are studied using a hierarchical structure. The first few layers deal with features that broadly classify the characters into small size groups. The lower level features are more specific to each character within a group. Higher level features can be identified based on pixel density and arrangement, while the lower level features have been identified using a chain code technique. The algorithm progresses successively through each group in the hierarchy until it finds a match with the input character.

Several Methods of Feature Extraction to Help in Optical Character Recognition

“…They used chain code histogram as feature on each of wavelet sub-bands in different resolutions, multi-layer perceptron (MLP) neural networks are trained for classification on each resolution, and then combined the classification results by majority voting technique. Sikdar et al [7] extracted the features of printed words by chain code method. Authors in [8] have summarized the recent trends in feature extraction and feature selection…”