Research on Style Transfer Method from Human Facial Images to Peking Opera Facial Images

Face makeup transfer is designed to transfer the makeup features of one reference image to another face image to achieve the transformation of different makeup styles. Ordinary makeup has the characteristics of uniform color in a single area and simple color in the overall area, while opera facial makeup is mostly hand-painted images, rich in colors and textures. It is difficult for non-professional actors to draw. At present, there is no deep learning-based method to study makeup transfer between face and Peking Opera facial makeup images. In this paper, we proposes a facial makeup transfer method based on generative adversarial networks and affine transformation. We create a facial makeup dataset from Peking Opera facial makeup images, and obtains the corresponding mask dataset through a semantic segmentation feature extraction model for facial parsing. A local histogram matching strategy is used to calculate the pixel-level histogram loss, which can better preserve the color characteristics of the image. An auxiliary discriminant network is introduced to synthesize reference labels by performing affine transformation on Peking Opera facial makeup images and face images, which helps the generator network to better reduce the information loss caused by image transformation. Through the research in this paper, it is expected to contribute to the research and application of this field, and provide new possibilities for the protection and inheritance of traditional culture and art.