Face Anonymization Made Simple
Han-Wei Kung, Tuomas Varanka, Sanjay Saha, Terence Sim, Nicu Sebe
2024-11-04
Summary
This paper presents a new method for anonymizing faces in images, making it easier to protect people's identities while maintaining high-quality visuals. The approach uses diffusion models to create detailed images without needing extra data like facial landmarks or masks.
What's the problem?
Current methods for face anonymization often rely on complex calculations from face recognition models, which can be inaccurate. They also typically require additional information, such as facial landmarks or masks, to guide the process, making them less efficient and harder to use.
What's the solution?
The authors propose a simpler method that uses diffusion models with just a reconstruction loss, which means it focuses on creating clear images without needing extra data. This method allows for high-quality face anonymization while also being able to perform face swapping tasks when given another facial image. They tested their model on public benchmarks and found that it performs better than existing techniques in preserving identity anonymity, maintaining facial features, and ensuring overall image quality.
Why it matters?
This research is significant because it provides an effective way to anonymize faces in images, which is crucial for protecting privacy in various applications like surveillance, social media, and research. By simplifying the process and improving the quality of the results, this method can help organizations comply with privacy regulations while still being able to analyze and use visual data.
Abstract
Current face anonymization techniques often depend on identity loss calculated by face recognition models, which can be inaccurate and unreliable. Additionally, many methods require supplementary data such as facial landmarks and masks to guide the synthesis process. In contrast, our approach uses diffusion models with only a reconstruction loss, eliminating the need for facial landmarks or masks while still producing images with intricate, fine-grained details. We validated our results on two public benchmarks through both quantitative and qualitative evaluations. Our model achieves state-of-the-art performance in three key areas: identity anonymization, facial attribute preservation, and image quality. Beyond its primary function of anonymization, our model can also perform face swapping tasks by incorporating an additional facial image as input, demonstrating its versatility and potential for diverse applications. Our code and models are available at https://github.com/hanweikung/face_anon_simple .