On Ethnic Pattern Generation Based On Generative Adversarial Networks

China is a unified multi-ethnic country,and ethnic patterns are an important part of ethnic culture and intangible cultural heritage with distinct ethnic characteristics.However,due to the insufficient technology of collection and the high threshold for innovative design of ethnic motifs,the research of ethnic motifs in academic and artistic circles is restricted,which is not conducive to the protection of ethnic motifs,and even less conducive to the inheritance and development of ethnic motifs.In this paper,we take Mongolian ethnic patterns as the research object and conduct a technical study on ethnic pattern generation and super-resolution reconstruction based on generative adversarial networks to address the problems of low quality of ethnic pattern data collection and lack of innovation.A database including 1621 Mongolian ethnic patterns is used in this paper,and the following work is done.The dataset is first preprocessed,then Style GAN2 is trained using the preprocessed images and the trained images are generated,and finally the generated images are super-resolution reconstructed using ESRGAN to generate indistinguishable,high-resolution patterns with Mongolian style for the human eye.In addition,DCGAN and SRGAN are used to compare with Style GAN2 and ESRGAN respectively.Compared with the time-consuming and labor-intensive traditional ethnic pattern design methods,this system lowers the threshold of ethnic pattern innovation and contributes to the innovative design of ethnic patterns to a certain extent,which has certain positive significance for the protection and inheritance of ethnic patterns.