Research of Image Defogging with Hierarchical Search and Texture Filter

In this paper, an improved image defogging algorithm with hierarchical search and texture filter is proposed. First, the hazed images are classified by whether the embrace sky area. Then, a hierarchical search based on subdivision and adaptive strategy for hazed images with sky or without sky is used to obtain the atmospheric light. For the image with sky, the atmospheric light is estimated according to the sky with low luminance value to obtain bright and clean sky without color distortion. For the image without sky, an SVM-based method is used to estimate reliable atmospheric light. For the transmission calculation, a precise method based on texture smooth and guide filter is proposed to enhance the contrast of the recovered images. Finally, a post-processed by the joint bilateral filter is present to suppress the noise in the area with low luminance. Testing result shows apparent improvement in the image defog that are bright and clean without local color distortion. And the performance of this algorithm using GPU parallel platform is demonstrated this algorithm achieves a frame rate of 24 fps for the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1024 \times 768$</tex> image, about 14 times faster than the CPU implementation. It is a feasible and practical solution.