Dense Hazy Image Enhancement Based on Generalized Imaging Model
Yuanyuan Gao, Guoliang Liu, Chao Ma · 2018
Image de-hazing is important for many computer vision applications. However, dense haze removal from a single image remains to be a challenging problem. Key insight that limits the performance of existing de-hazing algorithms is that these algorithms utilize the classic haze imaging model, which is based on an assumption that radiation on the object surface is sufficient and white. However, in dense hazy conditions, this hypothesis is easily broken. Thus, removing dense haze using classic de-hazing algorithms would result in dark-look or color shift. Therefore, in this paper, we propose a dense hazy image enhancement algorithm based on the generalized haze imaging model. The proposed algorithm includes two steps: Frist, we estimate pseudo ambient illumination and remove it to obtain an illumination balanced result. Second, we calculate the scene reflectivity as the enhanced result based on the spherical coordinate system. Experimental results demonstrate that the proposed algorithm surpasses state-of-the-art algorithms in most cases.