An Improved Estimation Method for Single Image Dehazing Model

Abstract

In this paper we present an improved method for estimating the optical transmission t in hazy scenes in a given single input image. Firstly, a new formulation to estimate t is created by combining constant albedo and dark channel prior. Secondly, the watershed segmentation is introduced to divide the whole areas into some gray level consistent parts according to color distribution in the image so that user can better estimate the atmospheric light A, and also further avoid halo artifacts phenomenon. Finally, 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 experiment results demonstrate that our proposed method can provide comparable results to dark channel prior and obtain reliable estimation value t with the advantage of minimal halo artifacts and fewer unreal details.