Image Dehazing

Image dehazing estimates a clearer scene from images obscured by atmospheric haze, fog, or smoke, supporting both visual use and downstream perception. Methods range from physical scattering models and hand-crafted priors to CNNs, Transformers, diffusion and flow models, and explicit scene representations; they learn from paired data or adapt without clear references. Evaluation spans synthetic and real-world scenes—including nighttime, aerial and remote-sensing, and video settings—using image-quality measures alongside detection, visibility, and novel-view reconstruction. A central challenge is transferring from simplified training haze to variable real conditions while preserving color and fine detail, maintaining physical and multi-view consistency, and meeting deployment costs.