A lightweight off-line handwritten Chinese character recognition algorithm based on gradient feature and channel attention

Handwritten Chinese character recognition has a broad application prospect in the field of optical character recognition system and handwritten character input device. In order to solve the problem that there are a large number of Chinese characters with various handwriting styles and plenty of similar characters, this paper proposes a lightweight off-line handwritten Chinese character recognition algorithm— Handwritten Chinese Character Recognition Depthwise Seperable Convolutional Neural Network (HCCR-DSCNN) based on gradient features and channel attention. Firstly, HCCR-DSCNN combines the traditional gradient feature extraction method to process the original image, and takes the eight-direction gradient feature as the input of the network, and uses the prior knowledge of Chinese character field to express the Chinese character image more effectively; then, the importance of each channel is obtained by introducing the channel attention mechanism to improve the recognition accuracy of the network. Finally, the traditional convolution is replaced by the depth-separable convolution to reduce the network parameters and computation. Experimental results show that the proposed algorithm can improve the recognition performance significantly on CASIA-HWDB dataset, and the accuracy of off-line handwritten Chinese character classification task reaches 94.16%.