Dehaze-GaussianImage: Zero-Shot Dehazing via Efficient 2D Gaussian Splatting Representation
2026-06-15 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors propose a new way to remove haze from single images by using a method called 2D Gaussian Splatting, which treats the image as a flexible field of Gaussian shapes instead of fixed pixels. Their approach uses a special learning strategy that separates the effects of haze from the clear parts of the image directly within this Gaussian model. This helps improve image clarity while reducing common problems like artifacts, all without needing any pre-trained data. Their results show that this method works very well with fewer parameters compared to existing approaches.
image dehazing2D Gaussian Splattingzero-shot learningatmospheric scattering modelanisotropic Gaussian fieldsreconstruction-decouplingunsupervised learningphysical priorsconvolutional neural networkstransformers
Authors
Yuhan Chen, Wenxuan Yu, Guofa Li, Kunyang Huang, Ying Fang, Yicui Shi, Wenbo Chu, Keqiang Li
Abstract
Existing single image dehazing methods are often constrained by computational redundancy in pixel-level optimization and the lack of physical interpretability in implicit neural networks. These limitations hinder the balance between representation efficiency and reconstruction fidelity. To address these issues, we propose Dehaze-GaussianImage, the first zero-shot framework that introduces 2D Gaussian Splatting (2DGS) into the image dehazing domain to break the traditional pixel-grid processing paradigm. Distinct from static convolutional neural networks (CNNs) or Transformers, our approach models hazy images as continuous and dynamically evolvable anisotropic Gaussian fields. Specifically, we propose a novel reconstruction-decoupling zero-shot learning strategy that embeds the atmospheric scattering model into the Gaussian parameter space. This strategy drives Gaussian primitives to adaptively split, clone, and prune during optimization, achieving geometric-level decoupling of the transmission medium and clear textures. Furthermore, explicit structure-preserving constraints are introduced to suppress artifacts commonly caused by traditional physical priors. Experimental results demonstrate that the proposed method achieves state-of-the-art (SOTA) performance in a fully unsupervised manner with minimal parameters, highlighting the potential of explicit Gaussian representation for low-level vision tasks.