Bangla Handwritten Single, Numeral, Vowel Modifier and Compound Characters Recognition Using CNN

Abstract

The difficulty of handwritten character identification varies by language, owing to differences in shapes, lines, numbers, and size of characters. There are several studies for the identification of handwritten characters accessible for English in comparison with other significant languages like Bangla. In their recognition procedures, existing technologies use multiple techniques such as classification tools and feature extraction. CNN has recently been shown to be proficient in handwritten character recognition in English. A Handwritten Bangla character identification system based on CNN has been examined in this research. Using CNN, the suggested approach for feature, labeling, and normalizing the handwritten character of images, as well as categorizing different characters. It doesn't use a feature extraction approach like previous research in the field. This research used almost 4,50,000 unique handwritten characters in a variety of styles. The recommended model has been proved to have a high recognition accuracy level and outperforms some of the most widely used methods already in use.