A New Method for Stroke Order Recognition of Handwritten Chinese Characters

Abstract

Correctly writing Chinese characters, which are composed of strokes, is an essential education content in the primary and secondary school in China. Automatically recognizing the strokes order is very useful to help teachers and students find the mistakes occurred in the handwriting process. In this paper, we propose a new method for recognizing the strokes order. We collect plenty of images of each stroke in the Chinese character writing process from the handwriting board, and use convolutional neural network (CNN) to build the classification model. To deal with the characters that are easy to be written in a wrong order, we add connection points in the case of the two connected adjacent strokes, and connection lines in the unconnected case. Thus, the information of the strokes order can be described more distinctly. Using this kind of extracted features, we can recognize these fallible characters more effectively. In the experiments, the real Chinese characters in the textbooks of primary school are used, and the results show the feasibility and efficiency of our proposed method.