Region-Aware Style Transfer Between Thangka Images via Combined Segmentation and Adaptive Style Fusion

Thangka is a unique intangible cultural heritage of the Tibet, with a long history and numerous schools, character-ized by distinctive techniques for depicting figures and expressing color. Style transfer is a popular research direction in computer vision, which can effectively extract the style from an image and incorporate it into another. Currently, most research on Thangka style transfer focuses on extracting Thangka styles and transferring them to non- Thangka images, while neglecting style transfer between different Thangka images. Style transfer between Thangka images presents two major challenges: first, the rich use of color and complex style in Thangka; second, directly applying style transfer does not adequately represent the original Thangka style in the generated image and can distort the content structure. Therefore, this paper proposes a method that combines segmentation and style transfer, enabling style transfer between corresponding regions of Thangka images. This approach contributes to the digital inheritance and preservation of Thangka art.