Abstract: Image degradation due to fog and haze presents significant challenges across numerous computer vision tasks, including autonomous navigation, remote sensing, and video surveillance. Traditional dehazing and defogging methods, often based on physical models and handcrafted priors, are limited in their adaptability to diverse and dynamic real-world conditions. With the rapid advancements in artificial intelligence (AI), particularly deep learning, a wide range of data-driven approaches have emerged, demonstrating superior performance in atmospheric image restoration. This survey provides a comprehensive review of recent progress in AI-based defogging and dehazing techniques. We systematically classify existing methods into supervised, semi-supervised, and unsupervised learning frameworks, examine popular network architectures, training strategies, loss functions, and benchmark datasets. Additionally, we analyse key evaluation metrics and compare the performance of leading approaches. The paper also discusses current challenges, such as generalization, real-time inference, and the scarcity of labeled data, while outlining promising directions for future research in AI-driven visibility enhancement.

Keywords: Fog and Haze Removal

| DOI: 10.17148/IARJSET.2025.12535