Applications Of Guided Filter In Fast Single Image Dehazing

Dehazing outdoor image, in other words, enhance the visibility of photos taken when fog and haze exists, is very important for improving performance of computer vision system which working on bad weather conditions. The atmospheric light model-based single image dehazing methods, as they simulate natural light conditions, have less restrictive conditions. Especially when dark channel prior is added, better results are received, which make it one of the hottest spots of image enhancement.Since the input is solely a low-contrast haze image, we need to estimate the depth map and the atmospheric light as well. Then we can receive the high-contrast image without haze. This is an ill-posed problem. The introduction of dark channel prior can help us with the estimation of depth map when the haze is heterogeneous. However when it is heterogeneous, it doesn’t work well. Traditional methods based on enhancement, although not restricted by haze distribution, may cause over-saturation effects due to the lack of depth information. Both methods involve optimization iterative process, which is quite time consuming.An image erosion is often necessary when operate the local contrast. Traditional methods rely on the optimization methods, which is often the most time consuming part of the process. We propose to use the edge-preserving guided filter to simulate this process approximately, which can achieve similar results and reduce the running time at the same time. Compared with the classical bilateral filtering, guided filter is an explicit filter, which is a linear-time algorithm handle the edges better.We combine the advantages of both methods here and proposed a method to enhance local contrast which is based on guided filter. Meanwhile, a post-processing method by tone mapping is proposed. This post-processing doesn’t involve any artificial light. By considering the depth map information, we determine the reference area of tone mapping, improve brightness of the overall image, and minimize the color distortion. Both simulation and quantitative assessment showed that, the proposed algorithm can deal with the heterogeneous fog, restore the visibility of the foggy image, and reduce the color distortion as well.