Degradation-Aware All-in-One Image Restoration via Latent Prior Encoding

2025-09-22

AI Summary

Real-world images are often degraded by spatially heterogeneous factors such as haze, rain, snow, and low illumination. Existing “one-for-all” image restoration methods rely on external textual prompts or handcrafted priors (e.g., frequency-based heuristics), limiting their generalizability. This paper proposes the first implicit prior learning framework for unified image restoration, modeling restoration as a degradation-aware inference process in latent space—enabling adaptive feature selection, spatial localization, and content reconstruction. Our approach jointly designs an implicit prior encoder and a lightweight decoder, augmented with an adaptive feature routing mechanism, eliminating dependence on explicit prompts or manual priors. Extensive experiments demonstrate state-of-the-art performance across six single-degradation, five composite-degradation, and unseen degradation scenarios, achieving an average PSNR gain of 1.68 dB and a threefold speedup in inference time.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Real-world images often suffer from spatially diverse degradations such as haze, rain, snow, and low-light, significantly impacting visual quality and downstream vision tasks. Existing all-in-one restoration (AIR) approaches either depend on external text prompts or embed hand-crafted architectural priors (e.g., frequency heuristics); both impose discrete, brittle assumptions that weaken generalization to unseen or mixed degradations. To address this limitation, we propose to reframe AIR as learned latent prior inference, where degradation-aware representations are automatically inferred from the input without explicit task cues. Based on latent priors, we formulate AIR as a structured reasoning paradigm: (1) which features to route (adaptive feature selection), (2) where to restore (spatial localization), and (3) what to restore (degradation semantics). We design a lightweight decoding module that efficiently leverages these latent encoded cues for spatially-adaptive restoration. Extensive experiments across six common degradation tasks, five compound settings, and previously unseen degradations demonstrate that our method outperforms state-of-the-art (SOTA) approaches, achieving an average PSNR improvement of 1.68 dB while being three times more efficient.

Problem

• Addressing spatially diverse image degradations like haze and low-light
• Overcoming limitations of text prompts and hand-crafted architectural priors
• Enabling adaptive restoration without explicit task cues via latent priors

Innovation

• Learned latent prior inference for degradation-aware representations
• Structured reasoning paradigm with adaptive feature routing
• Lightweight decoding module using spatially-adaptive restoration cues