Super-pixel based single image haze removal

Abstract

Single image dehazing is challenging because the number of freedoms is larger than the number of observations. Haze removal based on dark channel prior is effective, while it is an issue to determine the value of dark channel for each pixel. In this paper, a novel super-pixel based method is introduced to compute the value of dark channel for each pixel. A popular superpixel segmentation algorithm, simple linear iterative clustering (SLIC) is adopted to decompose an input haze image into super-pixels. After finding each super-pixel's dark channel, the dark channels for all the pixels in the super-pixel are set as the dark channel of the super-pixel. The transmission map is estimated from the dark channel and refined by the Guided Filter. In addition, an adaptive threshold is introduced for the transmission map to avoid amplifying noise in the sky region. Experiments show that our method has a good result with low computation.