The Reserch And Application Of Image Dehazing Algorithm

On the condition of fog and haze, the visibility of the scenery reduced greatly, whichcould lead to traffic paralysis, high-speed blockade and ship suspended. We need to find theright way to compensate for sight imitations in hazy weather. On the other hand, computeroutdoor surveillance systems, such as road monitoring, security monitoring, need clear input.In the hazy days, the image contrast is low and color fidelity is poor, where it threaten theaccuracy of intelligent identification and tracking, which are based on image feature extracted.Therefore, image dehazing has important practical significance and has received more andmore attention from scholars.We use image restoration method which is based on a physical model to make hazyimage clear in this paper. Base on the analysis of the causes of image degradation in hazy days.Then we use atmospheric scattering model to describe haze image imaging model, using darkchannel prior algorithm to obtain the relevant parameters of atmospheric scattering model，obtaining the real scenery information. This article mainly analyze three shortcomings of darkchannel prior algorithm in obtaining relevant parameters and then put forward relevantimprovement program.First, an adaptive window estimation method is proposed to solve the problem that theoriginal algorithm using constant window when we use minimum value filtering to get thedark channel value makes universality poor. The new method can select the appropriatefiltering window according to the image size.A modified scheme based on pixel transmission is proposed to solve the problem thatdark channel prior algorithm will cause color distortion when a large area of bright regionsexist. Mainly use the proximity between dark channel and atmospheric light, secondary use theproximity between the R, G, B channel to determining the condition as bright areas. For thebright regions, we use tolerance mechanisms to correct the error estimation of transmission,while non-bright regions are still used the original transmission estimation method.Original dark channel prior algorithm uses soft matting algorithm to refine transmissionfor removing the "halo" on the image and restoring the image detail. But it will cost a largenumber of time in solving sparse linear Laplace equation, which will make the algorithm inefficient. In this paper, the original matting algorithm is replaced by the guided filteringalgorithm. The guided filtering algorithm not only has good edge retention characteristics, butalso it’s just a Jacobi iteration based optimization matting algorithm, having linear timecomplexity.This paper combined three improvements to process a hazing image. Making imagedehazing range no longer limited, and improving the efficiency of dehazing at the same time.The algorithm is verified can work on any hazing image to restore its natural appearance withno color distortion by a large number of experiments.At last, the improved algorithm is firstly applied to snow blurred image, it perform wellin snow removal. Extending the existing application range of the dark channel prior algorithm.