Haze removal for single image based on physical model and guided filtering algorithm

Single image haze removal used median filtering algorithm to optimize the atmospheric veil and the filter´s edge preserving effect is not ideal by Tarel´s algorithm. Therefore the Tarel´s algorithm haze removal results contain halos in depth discontinuity regions. For solving this problem, some adjustments and improvements are made and an improving Tarel´s algorithm is proposed in this paper. Firstly we use the image edge detection method to locate the regions of depth discontinuity, then based on the detection map, we construct an improved guided filter which has larger smoothing factor in regions of depth discontinuity and smaller smoothing factor in regions of texture, and use the improved guided filter to optimize the atmospheric veil. So the optimized atmospheric veil can accurately reflect the depth information of the scene. The results of experiment show that our method can successfully recover the detail scene compared with Tarel´s algorithm. At the same time, it can also effectively weaken the generation of halos in regions of depth discontinuity, owing to guided filter´s low complexity, compared to the Tarel´s algorithm of single image haze removal, this method speed up the running process by 40 percent. So, it has certain values in both theory and practice.