Improving Dark Channel Prior for Single Image Dehazing

Abstract

This paper proposes an improved dark channel prior for removing haze from images. Dark channel prior is an effective method for removing haze. Dark channel is an image in the same size as the hazy image which is obtained by dividing the RGB images into windows and for each window, the minimum of each R, G and B channels are calculated. Then again the minimum of these three values is calculated and is replaced on all pixels in that window. For removing haze from images by dark channel prior, it is necessary to find transmission coefficient of haze and for this, airlight must be estimated. By having these factors, haze-free images can be restored. The dark channel prior method does not yield favorable results for some images, especially for those containing smooth regions. To overcome on this deficiency of the dark channel prior approach, the hazy image is initially segmented into smooth and non-smooth regions in this paper. Then for removing haze from smooth regions, the Gamma correction approach is used for contrast enhancement. Finally, for non-smooth regions, depending to the severity of haze, dark channel prior might be applied several times. The subjective and objective image quality assessments attest superiority of the proposed method compared to dark channel prior in haze removing.