Research And Application Of Chinese Calligraphy Style Transfer For Cultural Relics Restoration

In the inheritance of Chinese calligraphy works,some content is damaged due to natural or human factors.Because of works' abstract content and different styles,the restoration work has always been a difficult problem in the field of cultural heritage restoration.At present,it relies on the expert's skills to complete the restoration.The problems of low efficiency and large differences in quality have hindered the “rebirth” of precious calligraphy and stone carvings.This thesis proposes a Chinese calligraphy characters style transfer method based on deep learning neural network.Based on the image or three-dimensional point cloud of cultural relics,the digital image of calligraphy characters is extracted,and the style transfer framework based on the generative adversarial networks is established.In addition,a prototype system is designed and developed.Finally,the virtual restoration of the damaged characters of the cultural relics is completed.The main work of this paper is as follows:(1)A high-precision digital extraction method for calligraphy characters in stone carvings,calligraphy and paintings is proposed.For the characteristics of stone carving artifacts and digitalizing methods of calligraphy and paintings,The methods of extraction and preprocessing of the calligraphy characters for 3D models(stone carvings)and 2D images(Calligraphy and painting relics)are constructed,which provide data support for deep learning networks;(2)A calligraphy characters style transfer framework based on the Generative Adversarial Networks is proposed.After analyzing and summarizing the existing deep learning model,a Chinese calligraphy characters style transfer framework based on the GAN is constructed.And the goal of the Chinese calligraphy characters stylized generation is realized;(3)An improved generative adversarial network model is proposed.According to the characteristics of Chinese calligraphy characters in terms of overall structure,local shape and ink color change,the original distance measurement method of GAN is improved,and the objective function is added to constrain the whole training process.And network hyperparameter optimization and adjustment was performed to improve the quality of the network model for the generation of Chinese calligraphy characters;(4)A Chinese calligraphy characters generation and cultural relics virtual restoration prototype system for calligraphy works and stone carving was designed and developed.This system realizes the digitalizing generation of missing characters based on existing calligraphy works,the characters generation for special artists and the restoration of broken characters.The experimental results show that the system is reasonable in design,can effectively extract and transfer the calligraphy characters style information,and can meet the virtual repair needs of the calligraphy characters in the cultural relics;...