Enhanced Multi-Style Transfer Method based on VGG Network
Zheng Yong, Juanni Jiao, Fange Ye, Wei Li · 2023
In the domain of conventional image style transfer, remarkable progress has been achieved; nonetheless, certain issues persist. A large grayscale image with only one channel of brightness information lacks diverse color and texture features. This limits the variety in style transformation, resulting in unsatisfactory diversity of effects. To overcome this challenge, the present study introduces an enhanced multi-style transfer approach founded upon the VGG network. To achieve a greater variety of style transfers for single-channel grayscale images. This is accomplished by introducing a dual VGG network architecture coupled with the incorporation of the Gram matrix concept. Additionally, a multi-style fusion module is introduced to merge various style transfers for single-channel grayscale images. The experimental outcomes unequivocally demonstrate that the proposed approach effectively accomplishes manifold style transfers within single-channel grayscale images. Furthermore, apart from retaining the inherent content attributes, it bestows images with fresh visual allure. In comparison to conventional methods, the approach showcased in this study achieves significant advancements in both the quality and diversity of style transfer. Consequently, this innovative approach not only expands the scope of image style transfer applications but also paves the way for a novel potential in the realm of style transformation for single-channel grayscale images.