Hybrid Neural Network for Handwritten Mathematical Expression Recognition system

Abstract

The mathematical expression is an essential part of every domain they provide the mathematical explanation for one’s theory. The technological advancement in the domain of artificial intelligence has aided in various handwritten recognition systems such as handwritten mathematical expression recognition. Symbol recognition and structural analysis are two major obstacles in handwritten mathematical expression recognition. In this paper, we propose a hybrid neural network algorithm called validator, tracker, attention, and parser (VTAP). The hybrid neural network algorithms like CRNN is a blend of recurrent neural network (RNN) and convolutional neural network (CNN). It has shown better and more accurate outputs than the native CNN and RNN algorithms alone. CROHME dataset is used, which is the most widely used dataset. The recognition is divided into 4 parts validator, tracker, attention, and parser (VTAP). A tracker is equipped with a group of Bi-Directional Recurrent Network (BRNN) with the Gated Recurrent Unit (GRU). Succeeded by a tracker, the parser uses a GRU lead by guided hybrid attention. The accuracy and the time complexity of VTAP is compared with existing work Tracker, Attention and Parser (TAP), VTAP shows up to 92.2% of accuracy while TAP shows an accuracy of 89%.