Research On Haze Removal Using Dark Channel Prior

With the rapid development of economic and improving living standards, security is becoming increasingly important. As the most effective means of security work, video surveillance is widely used in public security, transportation, finance and other areas. However, in bad weather conditions, obtained surveillance video image contrast declines, color fades, and the scene information damaged, leading to failure of the video and bring risks. Therefore haze removal researching is badly needed.Work in this paper focused on de-haze technologies and did following researches:This article studies the atmospheric scattering model, analyzes the fog light propagation conditions and the reasons for degraded images. On this basis, it studies in depth of the image de-haze algorithm based on the dark channel prior, analyzes its strengths and weaknesses.According to the frequency character that fog mainly affects the low frequency components of image, the image defog algorithm which based on wavelet transform has been proposed. This algorithm significantly reduces the transmission rate to the fog during the optimization part of the computation, which can save more than 90% of the time consumption.Analysis the reasons that haze remains at the edge of scene, improved the structure of dark channel filter, Proposed a transmission rate estimating algorithm based on nine points filtering, and increased the computing efficiency substantially.As dark channel prior fails in bright areas, this paper proposed an improved haze removal method based on dark channel, correcting the errors in transmission rate map caused by bright pixels through introducing a tolerance mechanism. Experiments results show that the improved method can correctly estimate transmission map to remove haze and get clear images.According to character of atomization video, a transmission rate optimization method based on space-time smoothing has been proposed, which greatly reduces the computing memory requirements, the algorithm can be ported to embedded systems; using of Kalman filtering can enhance the accuracy of intensity estimates, suppress the interference which noise against, and eliminate the color transition of the video frames.