Adaptive Image Dehazing via Improving Dark Channel Prior

Abstract

The dark channel prior (DCP) technique is an effective method to enhance hazy images. Dark channel is an image with the same size as the hazy image which represents the haze severity in different places of the image. The DCP method suffers from two problems: it is incapable for removing haze from smooth regions, causing blocking effects on these areas; it cannot properly reduce a haze with a non-monotonic behavior. In this paper, an adaptive image dehazing method is proposed based on the DCP method to solve the problem of this method. In this method, to overcome the dark channel deficiency of the blocking effects, the dark channel is initially extracted. The hazy image is subsequently segmented into smooth and non-smooth regions. Regarding the smooth regions, the pixel values in the dark channel are reduced by dividing them with a rather great number. To solve the second problem, depending upon the haze severity, the haze removing technique is applied repeatedly until all the regions of the image are enhanced. Finally, the Gamma correction approach is used for contrast enhancement of the smooth regions. The performed subjective and objective comparison attest the superiority of the proposed method to the DCP one in removing the haze.