Semantic-aware deidentification generative adversarial networks for identity anonymization

Abstract Privacy protection in the computer vision field has attracted increasing attention. Generative adversarial network-based methods have been explored for identity anonymization, but they do not take into consideration semantic information of images, which may result unrealistic or flawed facial results. In this paper, we propose a Semantic-aware De-identification Adversarial Network (SDGAN) model anonymization. To retain expression effectively, extract image using edge-aware graph representation network to constraint position, shape and relationship generated key features. Then is injected generator together with randomly selected de-Identification. ensure generation quality realistic-looking results, adopt SPADE architecture improve ability conditional GAN. Meanwhile, design hybrid discriminator composed an analysis module, VGG-based perceptual loss function, contrastive enhance both ID A comparison state-of-the-art baselines demonstrates that our achieves significantly improved de-identification (De-ID) performance provides more reliable faces. Our code data are available on https://github.com/kimhyeongbok/SDGAN