Chinese Character Recognition Based on Residual Separable Convolutional Neural Network

Abstract

In this paper, we present an end-to-end deep learning system for Chinese character recognition. We applied the system to recognize ID card photographed by smart phone in natural environment. The quality of ID card images was affected by illumination and shadow. We developed a preprocessing procedure to enhance the images and then to detect and extract Chinese characters and numbers from the images. The end-to-end deep convolutional neural network (CNN)consists of deep separable convolutional (DSC)layers, batch normalization (BN)layers, ReLu activation layers and residual connections to mitigate over-fitting and gradient vanish. To train the network, we constructed a dataset containing the 3765 most frequent Chinese characters. We tested the proposed system on a self-constructed Chinese character dataset and achieved an accuracy of 92.73%.