4KDehazeFlow: Ultra-High-Definition Image Dehazing via Flow Matching

2025-11-12

AI Summary

To address poor scene adaptability, high computational cost, and color distortion in ultra-high-resolution image dehazing, this paper proposes a continuous optimization framework based on flow matching and haze-aware vector fields. The method models haze via a learnable, physics-inspired parametric 3D lookup table encoder; formulates an ordinary differential equation (ODE) dynamical system guided by a haze-aware vector field; and employs a fourth-order Runge–Kutta (RK4) solver to ensure iterative stability and suppress artifacts. End-to-end trained, the approach outperforms seven state-of-the-art methods across multiple benchmarks, achieving an average PSNR gain of 2.0 dB. It demonstrates particular superiority in dense-haze region recovery and color fidelity. Key contributions include: (1) a differentiable, physically grounded haze parameterization; (2) an ODE-based continuous-time dehazing dynamics driven by haze semantics; and (3) numerically stable integration via RK4, enabling robust optimization for ultra-high-resolution inputs.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Ultra-High-Definition (UHD) image dehazing faces challenges such as limited scene adaptability in prior-based methods and high computational complexity with color distortion in deep learning approaches. To address these issues, we propose 4KDehazeFlow, a novel method based on Flow Matching and the Haze-Aware vector field. This method models the dehazing process as a progressive optimization of continuous vector field flow, providing efficient data-driven adaptive nonlinear color transformation for high-quality dehazing. Specifically, our method has the following advantages: 1) 4KDehazeFlow is a general method compatible with various deep learning networks, without relying on any specific network architecture. 2) We propose a learnable 3D lookup table (LUT) that encodes haze transformation parameters into a compact 3D mapping matrix, enabling efficient inference through precomputed mappings. 3) We utilize a fourth-order Runge-Kutta (RK4) ordinary differential equation (ODE) solver to stably solve the dehazing flow field through an accurate step-by-step iterative method, effectively suppressing artifacts. Extensive experiments show that 4KDehazeFlow exceeds seven state-of-the-art methods. It delivers a 2dB PSNR increase and better performance in dense haze and color fidelity.

Problem

• Addresses Ultra-High-Definition image dehazing challenges
• Models dehazing as progressive vector field flow optimization
• Enables efficient adaptive color transformation for haze removal

Innovation

• Flow Matching models progressive vector field optimization
• Learnable 3D LUT encodes haze transformation parameters
• Fourth-order Runge-Kutta solver stabilizes dehazing flow field