Face Aging Simulation with Deep Convolutional Generative Adversarial Networks

Human face aging process simulated by computer has become a hot research issue of the computer vision field. In this paper we propose an improved face aging model based on Deep Convolutional Generative Adversarial Network (DCGAN). In this model, a given face is first mapped to a personal latent vector and age-conditional vector through two sub-encoders. Inputting these two vectors into generator, and then stable and photo-realistic face images are generated by preserving personalized face features and changing age condition. Perceptual similarity loss replace adversarial loss of Generative Adversarial Networks (GANs) as the objective function in this paper. Based on the existing face database, the experiment results demonstrate that face images synthesized by our method enjoys better authenticity and accuracy.