Research On Traffic Image Defogging Method Based On Multi-scale Analysis

In recent years,the city of fog weather,especially the haze weather frequent brings to the urban intelligent traffic control.Heavy fog weather traffic video image quality drop,lower contrast,target vehicle and monitoring information transformed analysis.How to effectively deal with the fog of traffic video image become an urgent problem of modern traffic management.At first,this paper made simple introduction to the fog images,and the existing algorithms of image to fog has carried on the induction and summary,find the effect of the image to fog algorithm is good or bad and selection of atmospheric light values have close relations,in the traditional algorithms of image to fog,tend to select the brightest pixel values in the image as the atmospheric light value,but at the same time can lead to inaccurate estimates,such as white objects or glare pixel brightness is the largest.In order to more accurately estimate the atmospheric light value,in this paper,the algorithm used the method of multi-scale analysis,multi-scale transform for image target in order to reduce the scene of atmospheric optical value estimate,the influence of the fog image three channels of low-frequency image maximum pixel averaging to reduce the interference,greatly improve the accuracy of the atmospheric optical estimate.First,the traditional dark passage to fog algorithm is not implemented for the detail of the image processing,at the same time cost on the processing time is longer,the original input image size and the size of the processing window will affect the process of time,and to the fog effect,in this paper,based on wavelet transform is proposed to improve dark passage to fog algorithm,after wavelet transform component describes the details of the high frequency information,low-pass component with low frequency information,respectively,for low frequency images do dark channel to deal with the fog,the high frequency image denoising do enhance,implement make improvements in frequency domain,enhance image details.And through multistage sampling image input method,and the adaptive setting the size of the processing window,on the one hand,reduces the computation of program,on the other hand also avoided because of processing window or big or small and lead to poor image smoothness,details such as fuzzy distortion.Second,in order to further improve the fog weather traffic video image contrast and definition,this paper proposes a combination of NSCT and MRF to fog algorithm,by not under sampling LP pyramid filter and the direction of the next sampling filter group to implement the Contourlet transform,the more accurate the fog image decomposition,a strong and the weak edge information of the image edge information,and then take advantage of MRF algorithm distribution of each pixel of the image tag to the fog,combining NSCT algorithm using the nonlinear mapping function correlation image pixels,NSCT coefficients of image transmission diagram,get more clear fog image enhancement,greatly reduces the loss of detail information risk,thereby increasing the robustness of the image to fog.