Recognizing and Understanding Handwritten Joint Letter in Bangla

Abstract

Recently, some work has been done on Bangla Handwriting character recognition. But many of them are only capable of recognizing Bangla digits, vowel or consonant character only. For building a model that can recognize Bangla handwriting, it is important to work with joint letter also. This “Recognizing & understanding handwritten joint letter in Bangla” is a project to develop a model to recognize and understand Bangla handwritten joint letters. This model will be capable of taking single or multiple images as input and then it will automatically recognize which joint letter is in the image. For this project I have collected the dataset from multiple sources. In this project, I have used transfer learning with deep learning for getting best result. Here I implemented some Deep Neural Network (DNN) and compared the results of them. Among all of them ResNet50 provides the best result. The result of the model showed 96.37% accuracy on testing dataset which is better than other existing works. Keywords — Convolutional Neural Network, Transfer Learning, Handwritten Joint Letter Recognition, Handwritten Joint Letters.