Color Image Dehazing Based On Non-local Variational Model

In haze weather,atmospheric visibility can be reduced,and the working efficiency of outdoor acquisition equipment will be greatly reduced,the quality of the images taken is seriously attenuated.Therefore,it is an important research content and hotspot in the field of computer vision that how to remove the interference of haze in the image by certain processing means,so as to obtain high-quality image,make the recovered image have satisfactory visual effect and obtain more effective image information.Aiming at the shortcomings of in dark channel prior theory algorithms,this thesis proposes two different image dehazing algorithms.Firstly,an image dehazing algorithm based on layer-by-layer search and fast guide filter is proposed.For the problem that the calculation of sky atmospheric illumination value is not accurate,uses the layer-by-layer search method to optimize it,and uses the fast guide filter to refine the transmission,finally,the image enhancement method based on nonlinear superposition processing is adopted to solve the problem that the image is generally dark after dehazing.The improved algorithm can get the image with better definition and lower processing time,and the problem of dark in some areas of the image is also improved,but this algorithm cannot effectively solve the problem of texture and edge preservation in images,this thesis combines with the dark channel prior and nonlocal variational model respective advantages,H-NL-LTV model is proposed,and uses the Alternating Direction multiplier Method algorithm(ADMM)and auxiliary variable is introduced to solve this model,the texture and edge features of the image are maintained to the greatest extent after dehazing.In order to verify the effectiveness of the dehazing algorithm proposed in this thesis,three classical algorithms which include Kimmel Retinex algorithm,MSR algorithm,He algorithm are selected,and another two algorithms which have been proposed in this thesis,the result dehazing images are compared and analyzed respectively from two aspects of subjective quality evaluation and objective quality evaluation.Through different quality evaluation indexes,the advantages of the two dehazing algorithms proposed in this thesis are verified,it is verified that the image dehazing algorithm based on layer-by-layer search method and fast guide filter can get clear and natural dehaze images,while H-NL-LTV model can not only achieve haze removal effectively,but also can maintain the feature of image texture and edge.