HANDWRITTEN CHINESE CHARACTER AESTHETIC GRADING BASED ON DEEP LEARNING

Zhuang Ziming

1

庄子明(1993-)，女，硕士，主要研究方向：模式识别与图像处理

Zhang Honggang

1

张洪刚(1974-)，男，副教授、博士生导师，主要研究方向：模式识别与图像处理

1、School of Information and Communication Engineering, Beijing University of Posts and Telecommunications,Beijing,100876

Abstract：In order to help young children improve the quality of Chinese writing and assist Chinese character writing teaching better, this paper explored the related techniques of handwritten Chinese character aesthetic grading. Combined with the latest deep learning theory, this paper proposed a handwritten Chinese character aesthetic grading method based on similarity retrieval strategy. According to the two conditions of Chinese characters and aesthetic scores, the different categories of similarity search are divided. The CNN features extracted from the handwritten Chinese character deep learning network are innovatively combined with the traditional structural features to improve the accuracy of handwritten Chinese characters grading. For the application scenarios and model training needs of this project, this paper collected the handwritten Chinese character image data set with the aesthetic scores - the XiaoXueBao data set for model training and experimental testing. The experimental results showed that the model and method of this paper have high effectiveness and practical value for the aesthetic grading of handwritten Chinese characters.