Fast Single-image Defogging

Bad weather conditions such as fog, haze, and dust often reduce the performance of outdoor cameras. To improve the effectiveness of surveillance and in-vehicle cameras under such conditions, we propose a method based on a dark channel prior for quickly defogging images. It first estimates the intensity of the atmospheric light by searching the sky area in the foggy image. Then it estimates the transmission map by refining a coarse map from a fine map. Finally, it produces a clearer image from the foggy image by using the estimated intensity and the transmission map. When implemented on a notebook PC with a graphics processing unit (GPU), it was able to process 50 images (720 × 480 pixels) per second. It can thus be used for real-time processing of surveillance and in-vehicle system images.