An efficient handwritten Bangla character recognition system using computer vision and natural language processing

Abstract

An efficient handwritten character recognition system for an alpha-syllabary language like Bangla has always been a challenging issue. Despite being in demand, the number of papers being conducted on this was very infrequent. Alongside computer vision, our paper proposes the idea of using grapheme segmentation approach to create an effective system for handwritten Bangla character recognition. The system profoundly deals with the image of handwritten Bangla characters to preprocess through Computer Vision. To achieve the efficiency, we have segregated each Bangla word into grapheme roots and diacritics, whether its simple or compound character. Through these segments, we compare the roots and diacritics individually with the given dataset to recognise the characters. Thus, this system is capable of coping up with the limitations that previous models have by recognising any handwritten Bangla characters efficiently with great accuracy of 0.98357, 0.98208 and 0.94325 for vowel diacritics, consonant diacritics and grapheme root respectively. For proper reconstructed grapheme representation as output, we have approached a reconstruction method with grapheme segmentation. Thus, the implementation of efficient handwritten character recognition was achieved by computer vision and grapheme approach from NLP.