Handwritten Bangla character recognition using a soft computing paradigm embedded in two pass approach

Abstract

The work presented here mainly involves recognition of handwritten Bangla characters using a soft computing paradigm embedded within the framework of previously developed Two pass approach. Bangla script, which has nearly 400 character symbols, is used by a vast majority of population in India and Bangladesh. Considering the large number of character classes, Two pass approach is chosen here with some important extensions. Typically, Two pass approach coarsely classifies an unknown sample pattern in a group of pattern classes in first-pass. In second pass, it makes finer classification of the sample by determining its membership to a class within the group. Two major aspects of extensions made with Two pass approach here are development of an algorithmic approach for grouping of various pattern classes needed for the second pass and a soft computing methodology for optimal selection of local regions of character images for extraction of features, specific to each grouping of pattern classes. To test the performances of the method, benchmark databases of handwritten Bangla characters are also developed. A methodology for reading filled in forms of handwritten Bangla characters is also presented here. The experimental results show significant improvement of recognition rates on handwritten Bangla characters compared to traditionally followed Single pass approach.