A novel fast haze removal technique for single image using image pyramid

Fast single image dehazing has been a challenging problem in many fields, such as computer vision and real-time applications. Recently, many dehazing algorithms have been proposed based on the dark channel prior (DCP). However, these algorithms aim to improve the refinement of the raw transmission map, while ignore the computational complexity of DCP itself. Therefore, this paper proposes a new technique for fast single image haze removal, which achieves a good tradeoff between the dehazing performance and the computational complexity. We first decompose the observed haze image into a coarse image and a detail image using Gaussian-Laplacian pyramid. Then, the coarse image is dehazed by Dark Channel Prior and Guided filter (GDCP). For the size of the coarse image is 1/4 of the original image, the computational complexity is reduced sufficiently. However, the recomposed image is blurred, since the detail image is still haze. So, we employ an unsharp filter to sharpen the blurred recomposed image. Experimental results show that the proposed dehazing technique effectively removes haze, and significantly reduces the computational complexity by 69.59% on average, compared with traditional GDCP algorithm.