Haze removal based on multiple scattering model with superpixel algorithm

• A multiple scattering model for hazy image formation is proposed.
• The atmosphere point spread function is estimated with shape-numeric reasonability.
• The superpixel method is employed to estimate the transmission for halo prevention.
• Sky detection with superpixel merging to reduce the color distortion.

An efficient method to remove haze from single image based on dark channel prior and the multiple scattering described by atmosphere point spread function (APSF) is proposed in this paper. By means of statistical analysis of the quality assessments for more than 100 images, we apply generalized Gaussian distribution to approximate APSF kernel in the image domain with parameter mapping from both shape similarity and numeric reasonability. Furthermore, a superpixel method is employed for estimating the transmission on the sky and non-sky region, in order to mitigate the halo artifact around the sharp edges and reduce color distortion in the sky region. Therefore, the haze-free images with abundant distinguished details and little halo artifacts can be finally restored by first applying deconvolution to the original hazy image. A series of experiments are additionally implemented to demonstrate the effectiveness and robustness of the proposed dehazing algorithm from both qualitative and quantitative comparisons with the state-of-art.