Offline Bangla handwritten character recognition

Abstract

Optical character recognition is an attractive subject for research work now days. In our study of this topic for Bangla Character domain, we have come around some methods of feature extraction and classification along with various preprocessing steps. We have exploited these techniques for their advantages and disadvantages. Among these, the methods that came to our attention because of their high accuracy are Stroke Extraction and encoding which is mainly mathematical, Chaincode extraction based on the direction of the points and Principal Component Analysis(PCA) of the character image. In our research, we have combined these methods and implemented the system in Matlab that gave us better accuracy than their individual implementation.