Face makeup transfer based on Generative Adversarial Network
Ziyun Zhao, Yiping Shi, Xiner Li, Shaojie Shi, Jin Liu · 2023
The goal of the face makeup transfer algorithm is to transfer the face makeup from the specified reference image to the non-makeup face in the target image while maintaining the face structure of the target image during the transfer process. Aiming at the problem that the currently proposed facial makeup transfer network is prone to deformation and the strength of makeup is uncontrollable, we propose an algorithm to realize facial makeup transfer by using the generation of confrontation network, adding the mask area corresponding to the fusion of attention and makeup. Firstly, the corresponding masks of lip makeup, eye makeup and base makeup are generated based on the detection of face key points, and then the masks are embedded with the features extracted by the encoder to generate the corresponding feature matrix to ensure the consistency of the non-makeup area before and after the transfer. Secondly, the global style features are generated by integrating the feature results through the attention module, and the high-quality makeup image is synthesized while maintaining the integrity of the face structure. Our method can not only keep the face structure unchanged during the transfer process, but also form a more natural face makeup transfer effect. The experimental results show that the method has better training results on the data set, and the generated facial makeup transfer image is better than PSGAN and BeautyGAN in terms of visual experience, and the strength of the transferred makeup can be adjusted by parameters.