A Complete Bengali OCR: A Novel Hybrid Approach to Handwritten Bengali Character Recognition

Abstract

A complete Bengali OCR is presented incorporating a novel hybrid approach to handwritten Bengali character recognition. The idea is to combine structural analysis and template matching techniques in order to recognise the handwritten Bengali characters. Handwritten Bengali characters are inherently cursive and there is an absence of well-defined strokes. In this approach, the character set has been separated into different distinct sub-classes based one some distinguishable structural features. Details of several approaches to detect these structural features are presented. Structural and syntactic features have been generated from the training samples to generate distinct character signatures. A match dictionary has been devised based on these character signatures, that helps in collecting multiple prototypes from the training samples. A revised form of continuity analysis is applied to match the test characters to the characters in the match dictionary. This complete OCR has been thoroughly implemented and tested and very promising results have been achieved in this direction.