Face aging synthesis with identity drift heatmap using generative adversarial networks

Abstract

This work aims to generate higher-quality, more accurate age-progressed images using Generative Adversarial Networks (GANs), while introducing Identity Drift Heatmap. In our research, we worked with various GAN models and eventually focused on StarGAN - Hybrid Residual Attention Block to generate the intended face image and identity drift heatmap. While previous research has successfully used GANs for image synthesis, limitations still exist, as most papers mostly focus on generating age-progressed images. In our approach, we can visualize how identitypreservation deviates or changes while aging. Face age synthesis is extensively used in forensics and law enforcement fields, where precise, accurate, and detailed face progression is crucial. To achieve this, we proposed a StarGAN model with Hybrid Residual Attention Block (HRAB) that ensures smoother and more accurate age progression. In addition, our model is also capable of handling diverse facial structures and various aging factors such as hair color, skin texture, wrinkles, while maintaining realistic outputs. With the implementation of Identity Drift Heatmap, our research can be instrumental in various domains, from law enforcement to digital entertainment.