A Review on CNN-Based Unsupervised Dehazing Methods
Subhash Chand Agrawal, Rajesh Kumar Tripathi · 2026
Restoring degraded images due to weather conditions, particularly unfavourable atmospheric effects that make it harder to see the details in outdoor images, is one of the hardest image processing tasks. Haze presents difficulties in many computer vision applications, highlighting the necessity of improving contrast and restoring view in blurry images. An extensive review of convolutional neural network (CNN)-based unsupervised dehazing techniques is presented in this paper. The technique of image dehazing, which enhances the contrast and visibility of hazy images, has attracted a lot of interest from researchers. The survey begins by going over the physical model, datasets, problems and difficulties, typical types of dehazing techniques, and evaluation metrics that are frequently employed in dehazing studies. Understanding the fundamental ideas and evaluating the effectiveness of dehazing require these elements. This paper then offers a critical analysis on the different deep learning methods applied to image dehazing. The major goal is to provide an intuitive grasp of the fundamental methods that have significantly advanced the removal of haze. The study uses a variety of baseline techniques in both quantitative and qualitative tests to set baselines and assess the effectiveness of these algorithms. Finally, the study identifies problems and difficulties that still need to be addressed in the dehazing field.