Handwritten Chinese Character Recognition Based on Convolutional Neural Networks and TrueType Font Template Matching

Convolutional neural networks(CNN) have made significant advances in offline handwritten Chinese character recognition, CNN recognize characters based on local feature but not making use of the overall topology of the character to improve recognition accuracy. In this paper, we propose a method for handwritten Chinese character recognition by using CNN combined with TrueType font template matching. Firstly, a trained CNN is used to recognize the handwritten Chinese character, the output Top-N are selected as the candidate characters, and then the handwritten Chinese character is linearly transformed and matched with the TrueType font templates of the candidate characters respectively, the best matching candidate character is selected as the result. The experiments show that the method can combine the speed of CNN and the accuracy of template matching effectively, has higher accuracy than conventional CNN, especially for specific types of handwritten Chinese characters recognition.