Degradation Of Image Restoration Algorithm Based On Dark Channel Prior Under The Haze Day Traffic Scenes

With the rapid development of advanced computer technology, digital image processing algorithms for outdoor surveillance systems, such as road surveillance, traffic monitoring and other areas have been widely used. But in the haze day environment, monitoring system acquired images are often low contrast, blurred clarity, color distortion, and have a great impact to public safety monitoring work. Therefore, the images obtained under the haze day taken quickly and effectively ways to removing haze has a very important significance. This paper analyzes the research status of the removing haze algorithms at home and abroad, and study deeply on the image restoration algorithm under the haze day traffic scenes. This article focuses mainly on atmospheric scattering model and the theory of dark channel prior, and does research from the following aspects:1) This paper studied the atmospheric scattering model, and fundamentally analysis the image degradation in haze days. On the basis, it deeply studied the theory of dark channel prior and the removing haze model based on that theory. By using MATLAB simulation, we found the algorithm’s disadvantages.2) Analysis of the dark channel prior on defogging white area failure, this paper proposed an improved algorithm, by introducing the adaptive parameter modified transmittance inaccurate as a result of the interference caused by the sky region pixels. Experiments proved that the proposed algorithm can correctly receive the transmission rate and get a clear picture of the haze-free.3) For the problem of using soft matting optimizing transmittance in removing haze based on dark channel prior leading to low efficiency, the article proposed optimizing the transmission based on block and based on local Wiener filtering. The algorithm of optimizing transmittance based on block abandons time-consuming by using block-level template optimizing for small localized area, without optimizing every pixels. This way can not only remove the edge effects, but also maintain the characteristics of gray edges, and the transmission image edges are not blurred. Meanwhile the algorithm improves the efficiency. The solution of the transmittance which is based on local Wiener filtering combines with dark channel prior to optimize atmospheric light dissipation function, and then get the refined transmittance. At length, the clear haze-free images can be obtained directly through the atmospheric scattering model.Combing the theoretical analysis and experimental demonstrations, the three kinds of removing haze algorithms based on dark channel prior in this paper can restore a clear haze-free image, and the efficiency of removing haze has significantly improved.