Research On Image Dehazing Algorithm Based On Dark Channel Prior And Fractional Multi-Variation Regularization

In the process of collecting images,the computer vision system is easily disturbed by the external environment.Especially in the haze weather,the collected outdoor images are often seriously degraded,mainly manifested in the loss of clarity,color distortion,blurred texture details,etc.,and the image with damaged features as input will inevitably affect the system’s subsequent analysis and understanding of the image,impairing the working performance of the visual system.Therefore,it is of great practical significance to study how to effectively reconstruct the original image from the foggy degraded image to improve the application performance and robustness of the vision system.Fractional calculus can improve the high-frequency components of the signal while retaining the low-frequency components of the signal nonlinearly.Applying this theory to image dehazing can make the edge of the image more prominent,while retaining the texture information of the image.At the same time,the prior information used by the single regularized image dehazing model is insufficient,and more regularized prior information needs to be incorporated to improve the image dehazing effect.To this end,two different image dehazing models are proposed:1.Use the fractional derivative to model the image,integrate the dark channel prior theory to modify the initial transmittance estimation method,and establish a single image dehazing model with dark channel and fractional multi-regularization constraints.First,according to the dark channel prior principle,the estimation method of the initial transmittance is modified to obtain the refined transmittance.Secondly,the fractional derivative is introduced into the variational image dehazing model,and a fractional multiregularization constraint variational model is established to optimize the initial transmittance.The first term of the model is the L2 fidelity term,which is used to measure the initial transmittance.The difference between the transmittance and the refined transmittance;the second term is the fractional regularization term,which uses the fractional gradient map of the degraded haze image as a guide map to describe the image edges and texture details;the third term is the total variational regularization term,to preserve image edges and suppress noise.Then,the Alternating Direction Multiplier Method(ADMM)is used to efficiently solve the model.Finally,a tolerance mechanism is introduced to further correct the transmittance of bright areas including the sky to restore a potentially clear image.The defoggy images were compared with the images already processed by He algorithm,Kimmel Retinex algorithm,Fang algorithm,and Meng algorithm.The experimental results show that the image dehazed by this method can not only effectively retain the texture details,but also better suppress the generation of blockiness and artifacts,and has strong robustness.2.Combined with dark channel prior theory and fractional calculus theory,an image dehazing model based on dark channel and fractional TVBH(Total Variation and Bounded Hessian)is established.First,the dark channel prior theory is used to obtain the initial transmittance map of the atmospheric light value of the foggy image.Secondly,a fractional derivative is introduced,and a fractional multi-regularization constraint variational model is constructed to describe the transmittance and atmospheric light values finely.Use fractional transmittance regular term and transmittance data term to optimize transmittance;use fractional TV(Total Variation)regular term and fractional BH(Bounded Hessian)regular term as the regular term of the dehazing image to balance The speckle effect and staircase effect are maintained while maintaining the edge and fine structure of the image;the data items of transmittance and atmospheric light value are introduced to the dehazed image,so that the dehazed image is closer to the original image.Finally,the model is solved using the split Bregman algorithm and fast Fourier transform.The defoggy images were compared with the images already processed by He algorithm,Kimmel Retinex algorithm,Fang algorithm,and Meng algorithm.The experimental results show that the image dehazed by this method can not only suppress the step effect and speckle effect,but also retain the texture details of the image,and at the same time make the dehazed image clearer and more natural.3.In order to demonstrate the process and effect of image dehazing better,a single image dehazing visualization interface is designed according to the GUI interface function provided by Matlab.The He algorithm,Kimmel Retinex algorithm,Fang algorithm,Meng algorithm and the two improved algorithms in this paper are encapsulated to make quantitative analysis of different algorithms more convenient.