Bengali Handwritten Character Transformation: Basic to Compound and Compound to Basic Using Convolutional Neural Network

摘要

Bengali handwritten character recognization technique faces numerous complexities because of the unusual shape and stroke of the letters. In Bengali scripts, two or more basic characters construct a compound character. So it is challenging for the children to understand how these compound characters have been structured. Here, we recommend a system that can turn these basic characters into a compound character. Also, transfer a compound character into the basic characters by combining deep learning and image processing principles. The identification is begun by preprocessing the image using several image processing techniques, and then a convolutional neural network model is used for the transformation of the characters. Furthermore, the MapReduce technique is introduced for finding the appropriate character. This experiment was done on a dataset of more than 300,000 images and received an 89.20% accuracy rate.