A Study Of Single Image Haze Removal Base On Dark Channel Prior

With the development of science and technology, digital images are essential to some common systems or devices, such as intelligent traffic system, remote sensing system, outdoor surveillance system, and camera equipment, etc. People rely more and more on digital images to obtain useful information in various domains of modern life. The haze weather appears frequently in recent years. Images which are obtained by digital imaging equipment in haze weather become blurred as the impact of haze particle scattering effect, and the information within the captured image will be attenuated. Haze weather influence on daily life is growing. In haze weather conditions, the digital imaging equipment can't get useful information in the fields of transportation, monitoring and remote sensing, etc. So the study on recovering a haze scene is very important.Several different haze removal methods,suchu as Retinex, Fattel, and Dark channel prior, were compared with simulation results. The simulations reveal that Dark channel prior had good effects on single image haze removal, but halo effect could occur around the edge. Dark channel prior have no effect on sky region which led to color distortion in haze removal image. Many studies have been made in refine transmission to improve haze removal quality. In this paper, we compare several common method of refine transmission, and present a new refine transmission method to improve halo effect. Halo effect area is determined by the difference between the estimate transmission and the transmission refined by guided filter. The transmission which is refined by guided filter in halo effect area replaced by the transmission refined by median filter. Using this method, the too high transmission in halo effect area will be refined. The refine transmission method present by this paper which combine the advantages of median filter algorithm and the guided filter algorithm. This method can avoid black spots caused by median filter algorithm and halo effect residues caused by guided filter algorithm. The transmission of sky region was refined twice to reduce the impact of color distortion. Canny edge detection method is used to detect edges in haze image, the edge image can transformed into binary image by closing operation as the ground scene is filled with edges, then the transmission of sky region can be separated by the binary image which is used as a template. The original transmissions of sky region are replaced by new transmissions which are expanded with non-linear method. The transmission image should be smoothed with guided filter after the transmissions of sky region are replaced. Tests showed that the method of refine transmission present by this paper can reduce the impact of halo effect around edges and color distortion in sky region, and a good defogging result was achieved.