Dehazing for images with large sky region

Abstract

This paper introduces a haze removal algorithm based on region decomposition and features fusion to overcome the challenges of the dark channel prior-based algorithm, such as block effect and color distortion. In our proposed method, an image is decomposed with the quad-tree method based on gradient and grayscale information to obtain the sky regions. These sky regions are used as the seed point for region-growing, which will segment the image into sky and non-sky regions. A Gaussian filter is applied for smoothing on the segmented image, which is then used as a weight map to optimize the transmission image in the dark channel prior algorithm. Finally, the haze-free images are obtained based on an atmospheric scattering model and color compensation. Our experimental results demonstrated that images restored using this algorithm are generally clear and natural, and the algorithm is especially suitable for hazy images with large sky regions.