Glyphs Morphing For Chinese Characters Technology Based On Generative Adversarial Networks

Chinese calligraphy images have received widespread attention in the field of computer vision due to their unique artistry and structure characteristics,and the emergence of generative adversarial network has made new progress in the study of glyphs morphing for Chinese characters.For the poor performance of the deep generative model to achieve endto-end glyphs morphing for Chinese characters,and the problem of overfitting to a certain extent,the research on the glyphs morphing for Chinese characters technology based on generative adversarial network was carried out.The main research works are as follows:(1)Based on the fact that sampling in prior distribution can overcome the generation of semi-finished calligraphy images,a glyphs morphing for Chinese characters method combining the sampling of the prior distribution with the generative adversarial network is proposed.This method automatically generates new skeletons of Chinese characters,utilizes a classifier to discriminate randomly generated Chinese character skeletons to improve the structural correctness,then introduces U-shaped generative adversarial network to optimize Chinese character skeletons into calligraphy images.Comparing the experimental results and visualized results indicate combining the sampling of prior distribution and the generative adversarial network can enhance the recognizability and diversity of the calligraphy image.(2)On the basis that the optimization of Chinese character skeleton transformation can extract refined Chinese features,a glyphs morphing for Chinese characters method combined with multi-segment generative adversarial network is proposed.The method automatically extracts Chinese character skeletons and transforms skeletons to guide the correctness of the topology structure.The Chinese character skeleton extraction uses dual discriminators to improve the structural correctness and matching of the generated skeletons;the Chinese character skeleton transformation adopts a cascade structure to ensure the correctness and diversity of the Chinese character topology of the generated calligraphy images.The comparative experiments and visualization results show that the combination of multi-segment generative adversarial network significantly improves the effect of Chinese character glyphs deformation.In short,the combination of prior distribution sampling and optimization of Chinese character skeleton transformation can solve the glyphs morphing for Chinese characters in complex scenes,and effectively improve the morphing effect.