Research And Application On Chinese Calligraphy Characters Generation Method Based On Generative Adversarial Networks

Chinese calligraphy is a traditional art with Chinese characters as the carrier and mainly written with brush,which has profound cultural connotation.The research on automatic generation of Chinese calligraphy has a wide range of applications in the fields of the establishment of Chinese calligraphy font,the restoration of ancient inscriptions and the digital protection and inheritance of cultural heritage.Early traditional font generation methods were mainly based on the idea of splitting and merging to generate new Chinese characters,but its generation effect was more dependent on the effect of splitting strokes or components extraction and usually required manual post-processing.In the existing font generation methods based on deep learning,it is usually regarded as an image to image translation problem.However,due to the lack of effective guided information,the existing methods still perform poorly in the calligraphy generation task.For example,the generated calligraphy characters have problems such as missing strokes or redundancy,and the methods usually require the use of paired datasets for training.In response to the above problems,this paper proposes an effective Chinese calligraphy characters generation method and apply it to the establishment of Zhu Xi’s Bang Shu font library by using contour information and region-aware attention to effectively guide Chinese calligraphy characters generation,and combining with generative adversarial networks.The main contributions of this paper are as follows:(1)In view of the poor performance of existing Chinese calligraphy characters generation methods,this paper proposes a Chinese calligraphy characters generation method based on contour and region-aware attention,called CRA-GAN for short,inspired by the "contour-region decomposition" of Chinese calligraphy characters,using the contour of calligraphy characters as its style prior information and introducing region-aware attention to capture its content information on the basis of cycle generative adversarial networks.In addition,an adaptive pre-deformation module is introduced to solve the problem of degradation of model performance due to the large glyph difference between the source and target fonts.(2)To verify the effectiveness of the proposed method,this paper firstly collects eight Chinese calligraphy datasets and conducts a series of experiments on these datasets.The experimental results show that the performance of the proposed method is significantly better than the existing methods,and the proposed contour guided information,region-aware attention and adaptive pre-deformation modules are very effective.(3)Aiming at the problem of the lack of Zhu Xi’s Bang Shu font library,this paper generates a set of high quality Zhu Xi’s Bang Shu font library including the commonly used Chinese characters based on the proposed method CRA-GAN and existing rubbings of Zhu Xi’s Bang Shu “Thousand Characters”.Compared with other methods,this paper generates a better result of Zhu Xi’s Bang Shu font library.In addition,this paper also applies the established Zhu Xi’s Bang Shu font library as important decorative elements to the design of cultural and creative products,resulting in a series of cultural and creative products with rich cultural connotation.The research of this paper will provide important technical support for the protection and utilization of the intangible cultural heritage such as the White Deer Hollow Academy.