Single image dehazing based on multi-label graph cuts

Haze blurs image information and reduces the visibility of objects in the image, which seriously affects the performance of computer vision applications in a hazy environment. We propose an improved dehazing model based on multi-label graph cuts. A hazy image is modeled as an undirected graph. The multi-label graph cuts algorithm divides the image into subregions according to the functions of brightness and saturation. A subregion is selected to estimate atmospheric light based on saturation. Under the similarity of transmission in the same subregion, transmission is estimated by the distance between the pixel and atmospheric light in RGB space. Finally, the transmission map is regularized to recover a haze-free image. Experiments in different scenarios demonstrate the effectiveness of the proposed method than the state-of-the-art methods.