Perception-based Image Denoising via Generative Compression

TL;DR

This paper introduces a perception-based image denoising framework using generative compression, improving perceptual quality while balancing distortion and rate, with theoretical guarantees and experimental validation.

Contribution

It proposes a novel generative compression approach for denoising, combining GAN and diffusion models, with theoretical analysis and practical performance improvements.

Findings

Enhanced perceptual quality over traditional methods.

Theoretical bounds on denoising error and decoding probability.

Competitive distortion performance on benchmarks.

Abstract

Image denoising aims to remove noise while preserving structural details and perceptual realism, yet distortion-driven methods often produce over-smoothed reconstructions, especially under strong noise and distribution shift. This paper proposes a generative compression framework for perception-based denoising, where restoration is achieved by reconstructing from entropy-coded latent representations that enforce low-complexity structure, while generative decoders recover realistic textures via perceptual measures such as learned perceptual image patch similarity (LPIPS) loss and Wasserstein distance. Two complementary instantiations are introduced: (i) a conditional Wasserstein GAN (WGAN)-based compression denoiser that explicitly controls the rate-distortion-perception (RDP) trade-off, and (ii) a conditional diffusion-based reconstruction strategy that performs iterative denoising guided by compressed latents. We further establish non-asymptotic guarantees for the compression-based maximum-likelihood denoiser under additive Gaussian noise, including bounds on reconstruction error and decoding error probability. Experiments on synthetic and real-noise benchmarks demonstrate consistent perceptual improvements while maintaining competitive distortion performance.