A modified dark channel prior for improved dehazing

Images recorded under tough environment often exhibit problems such as being too light, too dark or not having enough contrast because they are degraded by a number of phenomena like fog, haze, rain and snow. This makes it essential to create accurate, high quality imagery which truly represents the scene. Only then these images will be useful for applications like obstacle detection, person identification, driver assistance etc. One effective method for removing haze is the dark channel prior method which assumes a single value for the airlight for the entire input image. This assumption is not always true in the case of images of natural outdoor scenes. So in this paper, we assume that the airlight values are different in different portions of the input image. With this assumption, we are able to achieve better dehazing in local regions of the image.