Single Image Haze Removal Based on Global-Local Optimization for Depth Map

摘要

With the wide application of computer vision system, image haze removal has become a new challenge. A great number of image dehazing methods are proposed, which have varying degrees of dehazing effects and different shortcomings. The color attenuation prior for image haze removal presents a new way based on depth map estimation. The novel method performs well with little distortion and natural colors. This paper discusses the color attenuation prior for image haze removal and proposes the haze removal method based on global-local optimization for depth map. Regarding the halo artifacts in dehazing images, we combine the minimum filter and minimum-maximum filter to detect the potential are as of the halo artifacts and suppress them. For the case of the underestimation of depth information, we take advantage of the atmospheric light estimation to perform global optimization for final depth map. Experimental results demonstrate excellent performance of the proposed method.