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 in unrealistic or flawed facial results. In this paper, we propose a Semantic-aware De-identification Generative Adversarial Network (SDGAN) model for identity anonymization. To retain the facial expression effectively, we extract the facial semantic image using the edge-aware graph representation network to constraint the position, shape and relationship of generated facial key features. Then the semantic image is injected into the generator together with the randomly selected identity information for de-Identification. To ensure the generation quality and realistic-looking results, we adopt the SPADE architecture to improve the generation ability of conditional GAN. Meanwhile, we design a hybrid identity discriminator composed of an image quality analysis module, a VGG-based perceptual loss function, and a contrastive identity loss to enhance both the generation quality and ID anonymization. A comparison with the state-of-the-art baselines demonstrates that our model achieves significantly improved de-identification (De-ID) performance and provides more reliable and realistic-looking generated faces. Our code and data are available on https://github.com/kimhyeongbok/SDGAN

Text-Guided Image Manipulation via Generative Adversarial Network With Referring Image Segmentation-Based Guidance

“…The recent studies [26]- [28] have reported that imageto-image translation has practical applications such as image quality enhancement, image retrieval, and identity anonymization. With this trend, in the research field of image-to-image translation, various approaches represented by image inpainting [6], [7], image colorization [2], [8], and style transfer [9], [10], have been proposed.…”

Manipulation of Age Variation Using StyleGAN Inversion and Fine-Tuning

“…Age manipulation techniques should convincingly change the subject's appearance across different age variations while maintaining their unique facial identity. Furthermore, the recent development of generative adversarial networks (GANs) [1] has achieved great performance in traditional computer vision tasks [2]- [4], especially in effectively synthesizing high-quality results in image generation and editing applications. In practice, various age manipulation architectures are based on adversarial networks, that can be classified into conditional GAN-based methods, imageto-image (im2im) translation, and latent space transformation methods.…”

“…(1) The proposed framework achieves lifelong face manipulation while preserving the facial identity of the input image. (2) We propose a StyleGAN fine-tuning method to reconstruct specific details using loss functions based on multiple masks. (3) Our method is compatible with well-studied StyleGAN-based techniques, e.g., stylization and changing pose techniques.…”