Single Image Dehazing Using Non-local Total Generalized Variation
Renjie He, Xiucai Huang · 2019
Single image dehazing has been a challenging problem due to its ill-posed nature. In this paper, a novel single image dehazing approach is proposed to accurately model the transmission map and suppress artifacts in the recovered haze-free image. Firstly, a coarse transmission is estimated using the patch based haze-line model. After that, a non-local Total Generalized Variation regularization is introduced to refine the transmission while preserving the local smoothness property and depth discontinuities. In addition, a regularized optimization is proposed for recovering the scene radiance without bringing artifacts boosting. Compared with the state-of-the-art dehazing methods, both quantitative and qualitative experimental results indicate that the proposed method is capable of obtaining an accurate transmission map and a visually plausible dehazed image.