Multi-scale Decomposition Dehazing with Polarimetric Vision

In this paper, the problem of simultaneous image dehazing of near and far scenes in hazy weather is addressed. We propose the multi-scale decomposition dehazing with polarimetric vision (Pol-MSD) algorithm to solve this problem. First, the contrast limited adaptive histogram equalization method is presented to obtain the contrast enhanced polarization image as the channel of near scene. Meanwhile, based on stokes theory, the incomplete normalized degree of polarization image is obtained as the channel of far scene. Then, the dual channels of near and far scene are decomposed into the base layers and the detail layers. To fusion the base layers, we use sobel gradient map (SGM) as the fusion weight of the base layers, and we proposed a fusion rule for the base layers by taking account of the SGM. To preserve the texture details of near and far scenes in detail layer fusion, we design a novel weighted least squares optimization scheme to fuse the detail layers. To evaluate the proposed Pol-MSD, we create a polarized hazy image datasets. Experiment results demonstrate that the proposed algorithm has significant advantages in qualitative and quantitative evaluations compared with other state-of-the-art dehazing methods.