Online Chinese Handwriting Recognition with Time Sequence Information

Abstract

Handwriting recognition has been a heated topic over years. Due to the development of deep learning, a lot of research has been done to apply convolutional neural network (CNN) model to this task, which have achieved outstanding accuracy. Instead of focusing merely on CNN models, this article takes the features of Chinese handwritten character into consideration and manages to extract the information of the stroke order of the online handwriting recognition. To achieve this, two methods are proposed: (1) Design a two-branch model combining CNN and recurrent neural network (RNN);(2) Give a new channel division strategy. Also, the task of advanced prediction of the character which little research has been worked on is the key point. With the information of stroke order and some data augmentation strategy, methods proposed have achieved satisfying results.