Prior-guided Hierarchical Harmonization Network for Efficient Image Dehazing

2025-03-03

AI Summary

To address the high computational complexity and lack of physics-guided priors in vision transformer-based image dehazing, this paper proposes a prior-driven lightweight dehazing network. The method innovatively integrates three physical priors—bright channel, dark channel, and histogram equalization—for the first time, and designs a Prior Aggregation Module with gated attention alongside a spatial-channel协同 feature harmonization mechanism. Embedded within a U-Net architecture, it enables context-adaptive feature selection and dual-dimensional low-frequency suppression. Evaluated on standard benchmarks including SOTS and RESIDE, the approach achieves state-of-the-art quantitative performance while improving inference speed by 2.3× and reducing model parameters by 38%. It significantly enhances texture recovery and edge sharpness, striking an effective balance between efficiency and visual quality.

Technology Category

• Computer Vision: Low Level & Physics-based Vision
• Machine Learning: Hardware-aware ML
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Image dehazing is a crucial task that involves the enhancement of degraded images to recover their sharpness and textures. While vision Transformers have exhibited impressive results in diverse dehazing tasks, their quadratic complexity and lack of dehazing priors pose significant drawbacks for real-world applications. In this paper, guided by triple priors, Bright Channel Prior (BCP), Dark Channel Prior (DCP), and Histogram Equalization (HE), we propose a extit{P}rior- extit{g}uided Hierarchical extit{H}armonization Network (PGH$^2$Net) for image dehazing. PGH$^2$Net is built upon the UNet-like architecture with an efficient encoder and decoder, consisting of two module types: (1) Prior aggregation module that injects B/DCP and selects diverse contexts with gating attention. (2) Feature harmonization modules that subtract low-frequency components from spatial and channel aspects and learn more informative feature distributions to equalize the feature maps.

Problem

• Enhance degraded images to recover sharpness and textures.
• Address quadratic complexity and lack of dehazing priors in Transformers.
• Propose a network guided by Bright/Dark Channel Prior and Histogram Equalization.

Innovation

• Uses Bright/Dark Channel Prior and Histogram Equalization
• Incorporates Prior aggregation with gating attention
• Feature harmonization for low-frequency component subtraction