In order to solve the problem that the method of representing Chinese character features by image pixels cannot effectively represent the essential features of Chinese characters and has high space complexity, a feature extraction method for Chinese character images was proposed. The method mainly includes three parts: binarization of Chinese character image, skeleton extraction of Chinese character image, and feature extraction of Chinese character image. Binarization eliminates noise in the image and improves the accuracy of image feature extraction. Skeleton extraction retains important pixels in the image, eliminates Irrelevant pixels. Graph feature extraction combines Chinese character key points with graph data structures to represent Chinese character shape features. Experiments were carried out on five fonts of 3 908 commonly used Chinese characters. The results show that the method can correctly extract the graph features of Chinese characters with complex strokes and effectively represent the essential features of Chinese characters. The maximum number of Chinese characters with the same graph features of different fonts is 3 195, and the performance of the method is relatively stable. An average of 22.6 graph nodes can be used for each Chinese character, 19.1 edge representations, compared to using single-channel images to represent Chinese character features, can greatly reduce the space complexity.