The Research And Application Of Single Image Haze Removal Algorithm

In the condition of haze and fog climatic, the captured pictures are hard to apply to many computer vision systems, such as the traffic detection, target tracking and satellite imaging etc. Therefore, the technology of image defogging has become a hot spot in the field of image enhancement.In practical applications, the speed of defogging algorithm is a very important problem. The method of single image haze removal using dark channel prior has been widely concerned because it can recover haze-free image, but it is easy to appear halo effect. To suppress halo effect, they need to refine the estimated transmission by using soft matting method, but it influences the performance of the whole algorithm. The medium filter defogging algorithm is simple and rapid, but the recovered image may have black spots, shading and fail for heavy haze image. Therefore, this dissertation deeply studied the reason of image degradation in fog and haze, and made some improvement to the existing defogging algorithm and presented new ideas, and made a great deal of experiments. The works can be summarized as follows:(1) The reasons of halo effect and black spots are analyzed in detail, and combined the minimum filter with medium filter, we processed a new filter to refine the transmission-double-regional filter algorithm. This algorithm keeps rapidity of medium filter and improves the accuracy of transmission estimation of the scene, especially in the estimation of the mutation region of the scene depth trans mittance, at last, it suppresses the halo effect and shading effects successfully.(2) Since the dehazed image is not as bright as the atmospheric light, a pseudo-dehazed image based on the atmospheric scattering model is proposed to solve the over-defogging problem. The defect of over-defogging can be cut down by the pixels fusion between the original dehazed image and the pseudo-dehazed image. Experiments show that the dehazing images with this method have better sharpness, color restore degrees, even on the dense fog and color distortion.(3) Practically, the global atmosphere light value uncertain due to the images are captured in the difference times, locations and climates. By summarizing the characteristics of the one-dimension entropy and average gray value of input image, a method of self-adaptively regulating of the global atmospheric light is proposed. This method can achieve the optimal brightness of haze-free image, and increase the robustness. (4) The pictures of kiln flame images are blurred by dust, it seriously affected the flame image processing. According to the existing algorithm invalid for coal-fame, trichotomy filter is proposed to solve the problem, which can improve the estimation accuracy of the transmittance of the light images. Experiments show that the dehazing images with the proposed method is clear and natural, and have good enhancement characteristics for each region.