Introduction
Due to adverse weather conditions, haze, smoke, or mist is the most common problem in outdoor scenes. A hazy environment reduces atmospheric visibility and hinders the performance of many computer-vision based applications used for object tracking, vehicle surveillance, road traffic regulation, and navigation. The process of removing haze using a single image and compensating for the attenuated energy is known as single image dehazing. Retrieving haze-free results using a single image is an ill-posed and under-constrained problem due to the lack of information such as the Transmission-map and air-light contribution. This research project aims to develop novel priors and boundary constraints using statistical/physical properties or heuristic assumptions to forecast the unknown information required for dehazing. To this end, we aim to propose AI powered dehazing techniques to handle different illumination and haze conditions that may be useful for restoring visibility in aerial, terrestrial, and underwater imaging.
