Adaptive Transmission Compensation via Human Visual System for Robust Single Image Dehazing

Dark channel prior has been used widely in single image haze removal because of its simple implementation and satisfactory performance. However, it often suffers from halo artifacts or noise amplification, over-dark and over-saturation looking in some images containing heavy fog or large sky patches where dark channel prior is not established. To resolve these problems, this paper proposes a robust single dehazing algorithm via adaptive transmission compensation based on human visual system. The key contributions of this paper are made as follows: firstly, two boundary constraints on transmission map are deduced for the minimum intensity preservation and halo artifacts or noise suppression. Secondly, an improved HVS segmentation algorithm is employed to detect saturation areas in the input image. Finally, an adaptive transmission compensation method is present to remove fog in nonsaturation areas and suppress the halo artifacts or noise in saturation areas. Experimental results indicate that this proposed method can robustly improve the visibility of the foggy image in the changeling condition.