Fi-Gaussian: Frequency-Aware Implicit Gaussian Splatting for Single Image Dehazing

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

Summary

Fi-Gaussian is a novel frequency-aware implicit Gaussian splatting network designed for single image dehazing, addressing challenges like high-frequency detail loss and inaccurate physical scattering modeling. Unlike explicit rendering methods, Fi-Gaussian uses implicit Gaussian splatting to represent clear images as a continuous 2D feature space distribution. Its core frequency-aware module decouples low-frequency structural and high-frequency texture information in the frequency domain, then performs adaptive Gaussian aggregation with complex-valued weights to recover fine details. Additionally, a physics-driven scattering renormalization mechanism estimates the transmission map and atmospheric light, guided by implicit Gaussian priors. Experiments on multiple benchmark datasets show Fi-Gaussian achieves state-of-the-art quantitative performance and visually superior dehazed results, validating its effectiveness for low-level vision tasks.

Key takeaway

For Computer Vision Engineers developing advanced dehazing solutions, Fi-Gaussian offers a compelling new approach. You should consider integrating frequency-aware implicit Gaussian splatting to overcome limitations of explicit rendering, particularly for preserving high-frequency details and accurately modeling atmospheric scattering. This method's state-of-the-art performance suggests exploring implicit representations and physics-driven priors could significantly enhance your image restoration pipelines.

Key insights

Fi-Gaussian uses frequency-aware implicit Gaussian splatting and physics-driven renormalization for state-of-the-art single image dehazing.

Principles

• Implicit Gaussian splatting models continuous image distributions.
• Decouple low and high frequencies for detail recovery.
• Physics-driven priors guide dehazing parameter estimation.

Method

Fi-Gaussian employs implicit Gaussian splatting, a frequency-aware module with complex-valued weights for detail recovery, and a physics-driven scattering renormalization mechanism to estimate transmission and atmospheric light.

In practice

• Apply implicit Gaussian splatting to low-level vision.
• Enhance dehazing by separating frequency components.
• Integrate physics-driven priors for robust estimation.

Topics

• Single Image Dehazing
• Gaussian Splatting
• Frequency Domain Processing
• Implicit Neural Representations
• Image Restoration
• Computer Vision

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

Related on AIssential

• Dehaze-GaussianImage: Zero-Shot Dehazing via Efficient 2D Gaussian Splatting Representation — Dehaze-GaussianImage uses 2D Gaussian Splatting for zero-shot dehazing, modeling haze as evolvable anisotropic Gaussian fields.
• Supercharging Thermal Gaussian Splatting with Depth Estimation — Thermal-to-Depth Gaussian Splatting (TDg) uses thermal images and depth for faster, robust 3D scene representation.
• 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.
• EnerGS: Energy-Based Gaussian Splatting with Partial Geometric Priors — EnerGS uses a soft energy field from partial geometry to guide 3D Gaussian Splatting, improving reconstruction quality.
• Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting — LGTM enables 4K novel view synthesis by decoupling geometry from resolution using textured Gaussian primitives.
• Single Image Defogging Using a Fourth-Order Telegraph PDE Guided by Physical Haze Modeling — This paper introduces a novel hybrid defogging model that integrates a fourth-order nonlinear Telegraph Partial Differential Equation (PDE) with a physical haze formation model for single image defogging.

Open in AIssential →

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