Haze Removal Based On Non-local Image Dehazing

In recent years,the fog is more and more serious,which leads to the degeneration of outdoor access to images,how to outdoor for image restoration in the fog into clear images of high quality,has become the common goal of many researchers.This paper emphasizes on the non_local image to fog algorithm research,through detailed analysis of the proposed algorithm,for its single use color value index clustering inaccurate estimation of initial transmissivity and adjacent pixels with spatial similarity,haze line problem of inaccurate,to keep the image edge effect and improve the haze the sharpness of the image for the purpose,to the local image to fog algorithm was improved,better maintain image edge effect at the same time to the haze effect is better.The main work of this paper are as follows:(1)A non-local image dehaze algorithm combining color image gradient is proposed.The non-local image dehazing algorithm only uses the color pixel index to cluster,and the spatial similarity between adjacent pixels is not considered,which leads to the poor detail of image edge after the fog.In order to solve this problem,inspired by bilateral filtering,this paper proposes the combination of the color image gradient non_local image to haze,the method of adding color image gradient clustering process,calculate the new initial transmissivity,and to further maintain image edge effect,used in refining transmittance orientation filter,thus make the images more clearly to haze and better keep the edge details.(2)A non-local image defog algorithm based on dark channel is proposed.Non_local algorithms of image to fog in the use of haze line to estimate the initial transmissivity,haze line two endpoints are respectively calculated directly to two extreme value point of fog model,it is not about a few classes may not exist without the actual situation of fog image point.Considering the above problems,the first to use the value of dark channel to modify the fog endpoint value at the end of the line,and based on dark channel image to fog algorithm is sensitive to the salt and pepper noise,to deal with the issue of the salt and pepper noise,the median filtering has inhibitory effect to the change of noise,so the fog image median filter in the first place,to get dark after median filtering channel figure(MDCP),with MDCP finally to improve the non_local algorithms of image to fog haze line in the end of the value.Experiment respectively in natural outdoor image and image synthesis of fog on the class diagram,although to fogeffect on the vision to maintain the original effect,but from the perspective of objective structure similarity and peak signal-to-noise ratio contrast effect is better.(3)A single image defog system was designed.Using the GUI interface function of Matlab,a visual interface of image defog was designed,and a better display of image defog flow through visual interface.The single image defog system mainly consists of two parts,part of which is the realization process of image defog algorithm,namely,the main interface of fog removal;The other part is the attention to operating the system,the help interface.