Thangka Image Style Transfer Based on Improved CycleGAN
HuaFei Song, HuiYuan Tang, Wenjin Hu, XinYue Shi · 2022 7th International Conference on Intelligent Computing and Signal Processing (ICSP) · 2022
Aiming at the problems of artefacts, color loss, blurred outlines, etc. affecting the aesthetics of the overall artistic effect in the style transfer of thangka images, a style transfer method of CycleGAN digital thangka images based on attention mechanism is proposed. After extracting the style feature and content feature of the image, it is input into the generator based on the attention mechanism, and the attention mechanism is applied to assign probability distribution information to each area of the digital thangka image, so that the important style area can get more attention. The experimental results show that the method enhances the image structure and texture information, can highlight the style transfer effect of digital thangka images, and is beneficial to improve the integrity and artistry of the output image.