Variational Regularized Methods for Single Image Dehazing
时 间：2019-01-07 15:00-16:00
Outdoor images captured in poor weather conditions (e.g., fog or haze) commonly suffer from reduced contrast and visibility. To improve image quality, this talk will provide two variational regularized methods for single image dehazing. The first method proposes a two-phase transmission estimation framework. The dark channel prior (DCP)-based coarse transmission map is estimated in the first phase. The coarse map is then refined using the total generalized variation (TGV)-regularized variational method in the second phase. The proposed two-phase framework has the capacity of generating natural-looking transmission map leading to satisfactory dehazing results. The second method is essentially a unified second-order variational framework which aims to refine the depth map and restore the haze-free image, simultaneously. The proposed framework is able to preserve the important structures in depth map and latent image assisting in improving image contrast and visibility. Experiments on both synthetic and realistic images have been implemented to illustrate the satisfactory imaging performance of the proposed methods.