Advancements in Image Dehazing: A Comprehensive Study from Past to Present

Weather-induced atmospheric phenomena such as haze pose significant challenges in terrestrial photography, particularly in capturing distant scenes where adequate light penetration is crucial. Dehazing, the process of removing haze from images, has emerged as a critical technique in computational photography and computer vision, with applications spanning video surveillance, autonomous driving, and remote sensing. This survey paper comprehensively examines existing image dehazing methodologies, addressing both traditional and modern approaches. The survey delves into various image enhancement methods, such as histogram equalization and Retinex-based algorithms, along with filtering techniques and image restoration methods are explored, including polarization-based (using multiple images) and single image dehazing approaches. Prior-based and learning-based image dehazing techniques are discussed, and compared, highlighting the challenges and advancements in each approach. Learning-based methods, particularly those leveraging Convolutional Neural Networks (CNNs), show promise but their complexity, coupled with hardware limitations and associated costs, must be carefully considered, especially when prioritizing real-time implementation. Through a thorough evaluation of existing algorithms using large-scale benchmark datasets, this survey provides insights into the strengths, limitations, and future directions of image dehazing research. The evaluation encompasses various criteria, including full-reference and no-reference metrics, subjective evaluation, and task-driven assessment. Overall, this survey paper aims to guide researchers and practitioners in navigating the evolving landscape of image dehazing techniques.