Abstract Digital artwork reconstruction technology brings new opportunities for the preservation, dissemination display and utilization of art resources. This paper focuses on the digital reconstruction of art works, and firstly introduces the style migration model based on generative adversarial network, and proposes three style migration models based on generative adversarial network, namely Pix2Pix, CycleGAN and StarGAN. Based on the demand, CycleGAN is selected as the basis to propose a generative adversarial network based on asymmetric cyclic consistency structure to optimize the style migration algorithm. By introducing the cyclic consistency loss and saliency edge loss, it better promotes model optimization and improves the quality of the generated images to realize the digital reconstruction of art works. In the digital artwork art style migration experiments, the style migration algorithm SSIM and COSIN test values of this paper are 0.442 and 0.97 respectively, which are better than the compared CycleGAN, DaulGAN and Pix2pix algorithms. And in the application practice of reconstruction of digitized artworks, the digitized artworks reconstructed using the style migration algorithm of this paper obtain the average gaze time and average number of gaze times of the subjects to reach 9.78s and 20 times, respectively, which are higher than the traditional Pix2pix method as a comparison.
Pu Wei (Wed,) studied this question.