Fast Image Haze Removel Algorithm Base On Dark Channel Prior

In outdoors video surveillance system,once haze or fog appears,the quality of the image captured by the video surveillance system is easily to be affected.The haze image loses its color,contrast,and a large number of details,which seriously undermines the performance of the video surveillance system based on the image processing technology.In the recent years,although the research of image haze removal has made a great progress,it still cannot be applied to the actual situation.The Dark Channel Prior algorithm,which is based on single image,can achieve some valuable results to the haze image.However,the algorithm consumes too much time,so it cannot be applied to real video surveillance system.This paper aims to improve the recovered images and time complexity of Dark Channel Prior algorithm based on its own principle and research of haze image's degradation and characteristics.Meanwhile,we need to avoid the steep time complexity that the soft matting algorithm brings,or a step further,reduce the original time complexity,and allow the haze removal algorithm achieve similar or even better results.By doing so,we can have the real-time effect and apply it to practical field.The innovative points of this paper are as follows:(1)Improve the haze removal results of edged views of Dark Channel Prior algorithm.While the original algorithm estimate the transmission map,it assumes that the transmission in local patch is constant,which results in estimation error of the transmission of the pixels on the edges.Although soft matting algorithm can effectively improve the block artifacts in recovered images,it cannot accurately correct the transmission of the pixels on the edges.Based on the neighborhood similarity,we proposed a method to recalculate the dark channel value,and consequently,we can improve the edge pixels in the recovered images by only correcting the transmission values of them;(2)With replacing the soft matting algorithm in original Dark Channel Prior algorithm by guided filter,we not only eliminate the block artifacts in recovered images but also maintain a lower time complexity;(3)Simplify the formula of transmission calculation.Through the analysis of the estimate value of atmosphere light,we have found that in most cases the three channels' values of atmosphere light are quite similar.As a result,we replace the atmosphere light value by the average of its three channels' values,which simplifies the transmission formula and reduces once calculation of dark channel values;(4)Refine the formula of transmission calculation by adding a constant,which is simple but effective for promoting the brightness of recovered images.Experiments show that the proposed improved Dark Channel Prior haze removal algorithm can achieve the same or even better haze removal results,and meanwhile,obviously lower the time complexity.Accordingly,apply this improved haze removal algorithm to video surveillance system and we can gain greater applicable values.