Perception-based Image Denoising via Generative Compression

In three linesPaper proposes generative compression framework for perception-based image denoising. Two approaches: conditional WGAN-based denoiser explicitly controlling rate-distortion-perception trade-off, and conditional diffusion-based iterative reconstruction guided by compressed latents. Theoretical guarantees and perceptual improvements demonstrated on synthetic and real-noise benchmarks.

Summary generated by Claude — human-verified