A Style Image Confrontation Generation Network Based on Markov Random Field

Abstract

A style-transfer generates against network based on Markov random field is proposed in this paper. Based on the original image, a new image is generated by generate network, and then the error between the original image and the style image is calculated using the discriminant network and backward propagation to the generate network, high-quality style transfer images are generated through the continuous confrontation of the two networks. In the quantification of style loss and content loss, we have introduced Markov random field, which uses its limitation on the spatial layout to reduce the distorted distortion of the generated image and improve the quality of the generated image. Experiments show that the network can quickly generate high-quality style transition images in a short time.