A Face Aging Network Based on Conditional Cycle Loss and The Principle of Homology Continuity

Due to the broad application prospects in cross-age face recognition and various entertainment projects, face aging has attracted extensive attention from computer vision, graphics, psychology and other fields. In recent years, the method based on Generative Adversarial Networks(GANs) has achieved great success in generating high quality images. Among them, as a special case of GANs, Conditional GANs (CGANs) introduces prior information in the process of image generation to guide the generator to generate sample images with specific conditions. Inspired by CGANs, and in order to solve some problems in face aging research: (1) the accuracy of aging; (2) the aging effect is realistic and natural; (3) identity information is invariant, we propose a face aging simulation method based on conditional cycle loss and the principle of homology continuity. Different from the previous CGANs method for aging process simulation, we reconstruct the input face by using the synthesized aging face and the age label of the input face. By minimizing the reconstruction loss, the identity information during the process of face aging is maintained. At the same time, in order to ensure the accuracy of aging, we introduce an age classification network based on the Principle of Homology-Continuity, which is more consistent with the process of human “cognition”. The experimental results show that the proposed method generates pleasing face aging results, and significantly reduces the number of parameter.