Single Image Haze Removal: Comparative Studies with Advanced Matting Approaches

Abstract

It is widely known that the existence of haze in the atmosphere reduces the quality of images, which are captured by sensors of camera. The haze or fog removal, called dehazing or defogging, is usually completed according to the model of physical degradation, which requires the solution of poorly posed inverse problem. To alleviate the difficulty of the inverse problem, the dark channel prior (DCP) was previsouly proposed and received much attention. In this paper, a DCP-based dehazing of a single input image is proposed. Considering the concerns raised in matting technology, we have replaced it by more advanced technologies, which manage the task more appropriately. After a series of experiments we decided to combine two methods forming an estimated semi-automatic algorithm for finding the corresponding values. The results on various foggy images show the strength of the proposed new algorithm.