Pyramidal and segment-concurrent acceleration of dark channel prior algorithm

Dark channel prior algorithm performs excellently in haze removal problem, but it is time consuming. Unlike the computational mathematics based methods, we give image processing ways to accelerate the algorithm. We first use pyramidal method to accelerate the algorithm, but the number of hierarchies could be used is limited, and it will bring roughness and undesired effect. Then the segmentation method is shown in the paper with bad block phenomenon. Additional subjections are used to stitch each segment together. The new algorithm greatly reduces the complexity and allows concurrent computation. By analyzing the Matting Laplacian Matrix, we also give explanation to the phenomena emerged in the experiments.