Recognition of Bangla Handwritten Characters using Feature Combinations

The success of any character recognition system largely depends on how well the distinguishing features of different characters describe the classes of characters. This paper studies the effect of the combinations of three different feature sets (i.e. chain code histogram (CH), longest run (LR) and Gabor wavelet based (GW) feature) on Bangla handwritten character recognition in order to maximize the number of discriminative features among different character classes. Different combinations of the three feature sets, namely LR & GW, CH & GW, LR & CH, and LR, CH & GW tested on a standard database of Bangla characters revealed that the combination of LR and CH features yielded better recognition accuracy compared to the other cases. It was also observed that size of the feature vectors in the combination played a key role in the recognition process.