Unsupervised Image Dehazing Using Smooth Approximation of Dark Channel Prior
Vedran Stipetić, Sven Lončarić · 2022
In this paper we propose a new unsupervised deep learning method for single image dehazing. The method is based on a new loss function that incorporates a smooth approximation of the famous dark channel prior. The method is used to train a neural network and results are compared to state of the art results of supervised neural networks. Evaluation is done by comparing increase in object detection on dehazed images as well as by visual inspection of results of dehazing on natural images.