Single Image Dehazing Algorithm Based on Sky Region Segmentation

摘要

Great progress has been made in image dehazing recently, while the sky region processed by these algorithms tends to be degraded by serious noise and color distortion. Based on the in-depth analysis on the degradation, the paper proposes an improved dehazing algorithm which first segments the haze image into the sky and non-sky region, then estimates the coefficients of the two regions separately, followed by the combination of the regions with a refining step. Experiment results show that the proposed algorithm provides significant advantage in generating smoother and more natural sky region, while keeping the non-sky region almost the same as the state of the art algorithms.