Image Dehazing with Dark Channel Prior and Novel Estimation Model

Abstract

Single Image Dehazing technology is widely needed in many fields. In order to solve the problem, we propose an improved and modified framework for estimating the optical transmission t in hazy scenes in a given single input image. At first, a novel formulation to the t estimation is presented with the combination of constant albedo and dark channel prior knowledge. Later, we introduce the watershed segmentation methodology into the algorithm to separate the image into some gray level consistent parts based on the original image’s color distribution and feature difference. As a result, we could estimate the atmospheric light A better and avoid the important drawback of artifacts phenomenon. At last, through this effective estimation to t and A, the scene visibility is largely increased and the haze-free scene contrasts can be better recovered. The experimental analysis shows that compared with other state-of-the-art algorithms, our proposed algorithm can provide promising results to dark channel prior and get corresponding reliable estimation value t with the advantage of minimal halo artifacts and fewer unreal details. Our method is more effective and robust.