Retinex Algorithm on Changing Scales for Haze Removal with Depth Map

Abstract

In order to improve the traffic visibility in haze weathers, a Retinex algorithm based on the changing scale for haze removal with a depth map is proposed. It requires the haze image dark channel prior treatment to obtain the estimated depth map. Then it is according to the depth map to calculate Retinex scales for different parts of a hazy image. Finally a single scale Retinex transform is performed for each part of the image. Experimental results show that the algorithm can effectively improve the traffic visibility of hazy images without halo phenomena. Compared with the existing multi-scale Retinex algorithm MSR, it has the higher speed and better enhancement effect for the images that have greatly different scene depths.