Dehaze-GaussianImage: Zero-Shot Dehazing via Efficient 2D Gaussian Splatting Representation

2026-06-15 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

Dehaze-GaussianImage is a novel zero-shot framework for single image dehazing that introduces 2D Gaussian Splatting (2DGS) representation, moving beyond traditional pixel-grid processing. This approach models hazy images as continuous, dynamically evolvable anisotropic Gaussian fields, addressing limitations of computational redundancy in pixel-level optimization and lack of physical interpretability in implicit neural networks. The method employs a reconstruction-decoupling zero-shot learning strategy, embedding the atmospheric scattering model into the Gaussian parameter space. This allows Gaussian primitives to adaptively split, clone, and prune during optimization, achieving geometric-level decoupling of the transmission medium and clear textures. Explicit structure-preserving constraints are also integrated to suppress common artifacts. Experimental results indicate Dehaze-GaussianImage achieves state-of-the-art performance in a fully unsupervised manner with minimal parameters.

Key takeaway

For Computer Vision Engineers developing image restoration systems, Dehaze-GaussianImage offers a new paradigm. You should consider integrating 2D Gaussian Splatting for zero-shot dehazing, especially when seeking high fidelity with minimal parameters. This approach allows for geometric decoupling of haze and textures, potentially reducing computational overhead compared to traditional pixel-level methods. Explore its application to other low-level vision tasks to improve efficiency and performance.

Key insights

Dehaze-GaussianImage uses 2D Gaussian Splatting for zero-shot dehazing, modeling haze as evolvable anisotropic Gaussian fields.

Principles

• Hazy images can be modeled as dynamic Gaussian fields.
• Decouple transmission medium and textures geometrically.
• Integrate physical models into parameter space.

Method

The method employs a reconstruction-decoupling zero-shot learning strategy. It embeds the atmospheric scattering model into Gaussian parameter space, driving Gaussian primitives to adaptively split, clone, and prune during optimization.

In practice

• Apply 2DGS for efficient low-level vision tasks.
• Develop zero-shot dehazing solutions.
• Explore geometric-level decoupling for image restoration.

Topics

• 2D Gaussian Splatting
• Zero-Shot Dehazing
• Image Restoration
• Computer Vision
• Atmospheric Scattering Model

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• Fi-Gaussian: Frequency-Aware Implicit Gaussian Splatting for Single Image Dehazing — Fi-Gaussian uses frequency-aware implicit Gaussian splatting and physics-driven renormalization for state-of-the-art single image dehazing.
• Towards UAV Image Dehazing: A UAV Atmospheric Scattering Model, Benchmark, and Geometry-Aware Deep Unfolding Network — A new UAV atmospheric scattering model and deep unfolding network effectively dehaze UAV images by accounting for non-uniform haze geometry.
• EnerGS: Energy-Based Gaussian Splatting with Partial Geometric Priors — EnerGS uses an adaptive geometric energy field and decoupled optimization to improve 3DGS reconstruction in sparse, partially observed scenes.
• Voronoi-guided Bilateral 2D Gaussian Splatting for Arbitrary-Scale Hyperspectral Image Super-Resolution — GaussianHSI uses Voronoi-guided 2D Gaussian splatting for arbitrary-scale hyperspectral image super-resolution.
• Three-dimensional Damage Visualization of Civil Structures via Gaussian Splatting-enabled Digital Twins — Gaussian Splatting enhances 3D damage visualization in civil infrastructure digital twins, improving accuracy and efficiency.
• Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting — LGTM enables 4K novel view synthesis by decoupling geometry from resolution using textured Gaussian primitives.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.