Image Encryption and Compression Based on Reversed Diffusion Model

Yilin Guo,Jianhui Chang,Yuhuai Zhang,Jian Zhang,Siwei Ma
DOI: https://doi.org/10.1109/pcs60826.2024.10566414
2024-01-01
Abstract:Nowadays, as critical conduits of communication, the information security of images and videos is particularly important. The existing encryption techniques usually transform images into high-frequency content that resembles noise, pre-senting significant challenges in achieving efficient compression. This paper presents an innovative collaborative approach that integrates image encryption and compression using a reversed diffusion model. This method, by reversing the typical process of diffusion models, adeptly changes encrypted high-frequency content into a domain that is more amenable to compression. Leveraging the reversible nature of the Denoising Diffusion Implicit Models (DDIM), our framework ensures the high-fidelity restoration of information. Our experimental findings demonstrate that this approach not only effectively encrypts images but also compresses the encrypted high-frequency noise content, outperforming Video Versatile Coding (VVC) in compression performance.
What problem does this paper attempt to address?