FaceAnonym: Face Anonymization Model via Latent Space Mapping

Machine learning has become a key driver of technological development, but the rising demand for AI applications involving human interaction necessitates access to large databases of human image data. However, the use of large real-world image datasets, particularly those that contain faces, has given rise to valid privacy concerns. We examine the critical issue of anonymizing image datasets that contain facial information in this paper. We hope to strike a balance between the requirement for data-driven advancements and preserving people's right to privacy by addressing these issues. In this paper, we propose a new method named FaceAnonym that de-identifies facial images by projecting them onto the latent space of a GAN model. This allows us to preserve the important characteristics of the face, such as shape, expression, and luminance, while still obscuring the identity of the individual. Finally, our method has been shown to be more effective than other methods at de-identifying facial images. It is also fast and easy to use, making it a practical solution for de-identifying large datasets of facial images.