Abstract:
The requirement of Bangla character recognition has become one of the prime attentions among the current researchers due to the increase of automated systems and usage of hand held devices. This paper presents a novel approach to recognize handwritten bangla numerals and addresses a robust feature extraction scheme that spawns 23- dimensional features based on the numeral's structure and topology. The recognition scheme is organized with a proposed decision tree that has been justified using entropy calculation. Considering inconsistency in individuals' writing style and the presence of significant curves and loops, the proposed feature extraction method restricts to less dimensional features of each numeral from 14600 data samples. The recognition time of this scheme is much lower than the existing procedures, since the preprocessing tasks have been performed during down sampling operation. The quick response time 13.04 ms and the higher accuracy (96.82%) explore a new era for the proposed scheme of being implemented in Bangla numeral LeadPad toys.
