Abstract:Supervised dehazing models, trained on synthetic hazy-clean image pairs, often face a notable decline in performance when applied to real-world scenes. Consequently, CycleGAN-based unpaired dehazing methods are proposed to improve the model’s generalization. One successful approach among these methods involves decomposing the physical properties of the atmospheric scattering model (ASM). However, estimating physical properties individually from input images is difficult without supervised labels, which ignores the semantic consistency between different physical regions. We claim semantic region information can offer additional geometric spatial constraints for estimating physical properties, as natural images can be divided into regions with similar scene depths. Motivated by this, we propose a novel generalized and realistic unpaired image dehazing framework via region-aware physical constraints (RPC-Dehaze). Our approach utilizes fine-grained semantic region maps from the Segment Anything Model (SAM) in a specially designed region prompt enhancement module. This enables the dehazing and hazing cyclic networks to learn region-aware physical constraints, leading to accurate estimation of haze imaging physical properties. In contrast to existing unpaired methods that treat dehazing and hazing networks equally, we incorporate Retinex theory into the hazing network, allowing it to learn diverse illumination effects in different regions. We adaptively refine the Retinex-based illumination component, resulting in more realistic hazy images. To further facilitate unsupervised learning in our framework, we propose a physical consensual contrastive regularization to ensure compact representation constraints in the latent feature space. Extensive experiments on synthetic and real image datasets show our method surpasses state-of-the-art unpaired dehazing methods in both effectiveness and generalization capability.

Physically Disentangled Intra- and Inter-domain Adaptation for Varicolored Haze Removal

From Coarse to Fine: A Two-Stage Network for Dense Haze removal

Contrastive Haze-Aware Learning for Dynamic Remote Sensing Image Dehazing.

DADRnet: Cross-domain Image Dehazing Via Domain Adaptation and Disentangled Representation.

Domain Adaptation for Image Dehazing

Leveraging Central-Surrounding Receptive Fields for Single Image Dehazing

Single image mixed dehazing method based on numerical iterative model and DehazeNet

From Synthetic to Real: Image Dehazing Collaborating with Unlabeled Real Data

Bridging the Gap Between Haze Scenarios: A Unified Image Dehazing Model

Progressive residual learning for single image dehazing

Discrete Haze Level Dehazing Network

Adaptive Single Image Dehazing Using Joint Local-Global Illumination Adjustment

DEHRFormer: Real-time Transformer for Depth Estimation and Haze Removal from Varicolored Haze Scenes

Toward Generalized and Realistic Unpaired Image Dehazing Via Region-Aware Physical Constraints

Guided Real Image Dehazing using YCbCr Color Space

Learning Depth-Aware Decomposition for Single Image Dehazing

Color-Constrained Dehazing Model

Single image dehazing via cycle-consistent adversarial networks with a multi-scale hybrid encoder-decoder and global correlation loss

Towards Domain Invariant Single Image Dehazing

Adaptive Haze Removal for Single Remote Sensing Image.

Multi-Scale Adaptive Dehazing Network.