Perception-based Image Denoising via Generative Compression

2026-02-12

AI Summary

This work addresses the challenge that existing image denoising methods often over-smooth images under strong noise and distribution shifts, struggling to balance structural preservation and perceptual fidelity. To this end, the authors propose a perception-oriented denoising approach grounded in a generative compression framework. Specifically, they design a conditional Wasserstein GAN-driven compressive denoiser to explicitly control the rate–distortion–perception trade-off, and further introduce a compressed latent-guided conditional diffusion model for iterative texture refinement. Theoretical analysis provides non-asymptotic performance guarantees for the compressed maximum likelihood denoiser under Gaussian noise. Experiments demonstrate that the proposed method significantly improves perceptual quality on both synthetic and real-world noise benchmarks while maintaining competitive distortion performance.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning on the Edge & Model Compression
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

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.

Problem

• image denoising
• perceptual realism
• distortion-perception trade-off
• generative compression
• distribution shift

Innovation

• generative compression
• perception-based denoising
• rate-distortion-perception trade-off
• conditional diffusion
• Wasserstein GAN