A Method for Style Transfer from Artistic Images Based on Depth Extraction Generative Adversarial Network

Abstract

Depth extraction generative adversarial network (DE-GAN) is designed for artistic work style transfer. Traditional style transfer models focus on extracting texture features and color features from style images through an autoencoding network by mixing texture features and color features using high-dimensional coding. In the aesthetics of artworks, the color, texture, shape, and spatial features of the artistic object together constitute the artistic style of the work. In this paper, we propose a multi-feature extractor to extract color features, texture features, depth features, and shape masks from style images with U-net, multi-factor extractor, fast Fourier transform, and MiDas depth estimation network. At the same time, a self-encoder structure is used as the content extraction network core to generate a network that shares style parameters with the feature extraction network and finally realizes the generation of artwork images in three-dimensional artistic styles. The experimental analysis shows that compared with other advanced methods, DE-GAN-generated images have higher subjective image quality, and the generated style pictures are more consistent with the aesthetic characteristics of real works of art. The quantitative data analysis shows that images generated using the DE-GAN method have better performance in terms of structural features, image distortion, image clarity, and texture details.