Asniyani Nur Haidar Binti Abdullah, Mohd Shafry Mohd Rahim, Sim Hiew Moi, Mohd Hafizan Musa, Ahmad Hoirul Basori, Novanto Yudistira, Dong-Lin Chen
Haze density estimation in single-image dehazing presents a fundamental challenge for computer vision systems operating under atmospheric degradation. This comprehensive review structures contemporary approaches around three critical challenges: (1) nonuniform haze distribution, (2) spectral distortions (particularly the characteristic bluish tint in distant scenes), and (3) the absence of standardized evaluation metrics exacerbated by limited benchmark datasets. Departing from conventional surveys, we establish a principled taxonomy of three dominant paradigms-physical prior-based methods, deep learning architectures, and hybrid optimization frameworks-systematically evaluating their effectiveness against spatial variance in haze density, wavelength-dependent scattering effects, and the lack of ground-truth references. Our analysis highlights significant advancements in attention mechanisms, physics-informed neural networks, and self-supervised learning strategies while revealing persistent gaps in real-time processing, spectral fidelity preservation (especially for bluish tint correction), and cross-environment generalization. Beyond quantitative benchmarks, this work provides researchers with both a structured understanding of current techniques and clear future directions, particularly in adaptive scattering modelling, computationally efficient deployment for practical vision systems, and the development of perceptually consistent evaluation frameworks that address the unique requirements of haze density estimation. This review systematically examines recent advances, compares methodological paradigms, and outlines key research gaps in datasets, spectral distortion correction, and learning-based generalization, offering actionable directions for developing perceptually faithful and computationally efficient dehazing systems. © 2026, Malque Publishing. All rights reserved.
Pervasive Computing and Educational Technology, Department of Media Interactive, Faculty of Information and Communication Technology, Universiti Teknikal Malaysia Melaka, Melaka, Malaysia; Department of Emergent Computing, Faculty of Computing, Universiti Teknologi Malaysia, Johor, Johor Bahru, Malaysia; Fakulti Sains Komputer dan Matematik, Universiti Teknologi MARA (Segamat), Johor, Malaysia; Department of Information Technology, Faculty of Computing and Information Technology in Rabigh, King Abdulaziz University, Rabigh, Saudi Arabia; Departemen Teknik Informatika, Fakultas Ilmu Komputer, Universitas Brawijaya, East Java, Malang, Indonesia; School of Big Data Science, Jiangxi Institute of Fashion Technology, Nanchang, China
