Image Dehazing Based On Dark Primary Color Prior And Second-order Variational Model

Traditional haze images contain not only haze,but also severe noise.Some of these noises are caused by small particles such as smoke and dust contained in the air,and some are due to the noise generated during the transmission of digital images taken on hazy days.The dehazing algorithm based on the dark primary color a priori can dehaze hazy images in different scenes,but the dehazed image usually contains noise and the partial detail preservation effect is not good.The second-order variational model can effectively remove noise,can preserve edges,and maintain texture details,and the second-order differential is more suitable for describing the oscillating nature of the image than the first-order differential,and can better suppress the staircase effect.Aiming at the problems in the dark primary color prior algorithm after dehazing,combined with the advantages of the second-order variational model,this paper first uses the dark primary color prior method to estimate the atmospheric light value and rough transmittance map of the foggy image,and then The linear diffusion model is applied to the solution of the fine transmittance map,and then it is combined with the second-order variation model Laplacian variation model,Hessian matrix variation model,total generalized variation model,total curvature variation model,Four second-order dehazing models are proposed: the dehazing model(H-LV)based on the dark primary color prior and the Laplace variational model,and the dehazing model(H-HMV)based on the dark primary color prior and the Hessian matrix variational model,the dehazing model(H-TGV)based on the dark primary color prior and total generalized variational model,the dehazing model(H-TCV)based on the dark primary color prior and total curvature variational model.In order to improve the calculation efficiency,the corresponding exchange direction multiplier solving algorithm is designed for the four models.In order to verify the dehazing effect of the four dehazing algorithms mentioned in this article,the images of the LIVE Image Defogging image database are randomly selected,and the proposed models and algorithms are experimentally verified.The classic algorithm is compared with the four algorithms proposed in this paper,and the dehazing images are compared and analyzed from two aspects: subjective quality evaluation and objective quality evaluation.Through the comparison of different quality evaluation indicators,the effectiveness of the four dehazing algorithms proposed in this paper is verified.The image edges obtained by dehazing are maintained well,and the image noise can be suppressed,and the quality of the four algorithms is judged to be good or bad.The experiment of keeping detail of inflection point for four algorithms was carried out.The experiment proves that the effect of keeping inflection point and detail of H-TGV and H-TCV is better.