Design of Online Handwritten Mathematical Expression Recognition System Based on Gated Recurrent Unit Recurrent Neural Network

Abstract

Handwritten input is an important input method for mobile devices. After developing for a long time, the technology of handwritten recognition is becoming more and more mature. However, the task of handwritten mathematical expression recognition is more complicated than general handwriting text recognition task. The large number of mathematical and alphabetic symbols and the complicated two-dimensional spatial structure increase the difficulty of handwritten mathematical expression recognition. In this paper, a system of integrated recognition without using grammar constraints was proposed for solving the problem of online handwritten mathematical expression recognition. In the proposed system, the sequences of features in stroke level are extracted from handwritten mathematical expression in chronological order, then the sequences of features are processed and recognized by the pre-training Gated Recurrent Unit (GRU) model with Connectionist Temporal Classification (CTC). Compared with the traditional grammar-driven recognition system of handwriting mathematical expression, the proposed system by passes accurate segmentation of mathematical expression sequence and the manual work for defining grammars. Therefore, the proposed system can reduce the time complexity of recognition algorithm of handwritten mathematical expression and improve the efficiency of recognition.