Application of Machine Learning Algorithm in Art Field – Taking Oil Painting as an Example

Abstract

AbstractOil painting production is a very time-consuming task. This article uses the current generation confrontation network popular in machine learning to transfer the style of images, and directly convert real-world images into high-quality oil paintings. In view of the current popular AnimeGAN and CartoonGAN generative confrontation networks, there are problems such as serious loss of details and color distortion in image migration. In this paper, by introducing SE-Residual Block (squeeze excitation residual block), comic face detection mechanism and optimizing the loss function, a new BicycleGAN is proposed to solve the problem of serious loss of details in the AnimeGAN migration image. By adding DSConv (distributed offset convolution), SceneryGAN is proposed to speed up the training speed and eliminate the ambiguous pixel blocks in the CartoonGAN migration image. The experimental results show that compared with AnimeGAN and CartoonGAN, the method in this paper has a significant improvement in training speed, comic image generation quality, and image local realism.KeywordsImage style transferGenerative confrontation networkAnimeGANCartoonGAN