Research on Face Age Synthesis Method

Face age synthesis technology simulates the natural change process of the human face as it grows or decreases with age by generating a model, which is of great value in practical applications such as cross-age identification and missing person search. This paper presents a systematic review of research advances in face age synthesis. In the current research context, methods for face age synthesis can be classified into classical methods and deep learning methods. Early classical methods can be divided into physical model-based methods, prototype-based methods and reconstruction-based methods; with the development of deep learning, the current stage of methods is centered on Generative Adversarial Networks (GAN) and their variants. Conditional GAN (cGAN) improves generative controllability through target age labels, disentangled GAN separates shape, texture and identity features, and Attention-based GAN and StyleGAN family of models can further optimize detail fidelity. However, identity drift in long-span synthesis, insufficient generalization across datasets and high computational cost remain key challenges. In the future, lightweight architectures, physiological a priori fusion and ethical compliance mechanisms need to be explored to promote applications in security, medical, and other scenarios.