Single image dehazing method via sky-regions segmentation and dark channel prior

Images and videos captured in hazy weather are often suffered from visual quality degradation. Recently many dehazing algorithms based on the dark channel prior theory have been proposed. However, these algorithms fail to achieve good performance if there are large sky regions or point light sources in the imaging scene. This paper proposes a novel dehazing algorithm based on sky region correction to overcome the problem of the dark channel prior which is not reasonable for sky regions. In order to estimate the atmospheric light, we choose the luminance value at the top 1% pixels of the sky area in the transmission map and calculate the average value of brightness by these pixels corresponding to the input image, by which the atmospheric light can be estimated more accurately. Experimental results demonstrate that the proposed algorithm has better performance in haze removing and contrast-enhancing than other methods.