Sky-Preserved Image Dehazing and Enhancement for Outdoor Scenes

Abstract

Outdoor imaging systems are vulnerable to severe weather including haze and fog. Image dehazing methods can be used to improve image quality without replacing cameras or using additional devices. However, it is not easy to effectively remove haze from outdoor images while preserving color fidelity and reducing artifacts, especially for sky regions due to their large depth of field. To solve these problems, we propose a novel sky-preserved image dehazing and enhancement approach based on region segmentation. First, regions of the sky and white objects are separated efficiently from other areas. Second, contrast limited adaptive histogram equalization (CLAHE) is processed for the sky region and multi-scale Retinex with color restoration (MSRCR) is carried out for non-sky and non-white areas, respectively. Finally, a region-ratio-based adaptive Gamma transformation is performed to correct the overall darkness and color distortions introduced by the original MSRCR. Experiments on diverse real-world hazy outdoor images show that the proposed approach outperforms the traditional MSRCR algorithm as well as three other state-of-the-art methods, in terms of contrast, image entropy, average gradient, and visual effects.