Detecting Handwritten Bengali Mathematical Expressions Using Neural Networks

Abstract

The problem of handwritten numerical expression recognition in languages such as Bengali is not well studied in machine learning, even though research on handwritten equations is well known. This work presents a thorough method for correctly identifying and analyzing handwritten Bengali mathematical statements using deep learning. Handwritten numerals and operators in Bengali are complicated, therefore models like custom CNN Layer, ResNet34, LetNet5, Xception and Vgg19. To guarantee robustness and dependability, the system's performance is assessed using measures including accuracy, precision, recall, and F1-score. The objective is to further the field of optical character recognition (OCR) for mathematical material in Bengali by creating an accurate and efficient recognition system.