Research On Single Image Haze Removal Algorithm Based On Adaptive Transmission And Retinex Theory

In recent years,with the development of computer technology,computer vision has gradually received people's attention,and it is widely used in artificial intelligence,geographic survey,traffic monitoring and military monitoring.However,hazy weather can cause severe degradation of images captured by outdoor computer vision systems,resulting in lower saturation,lower clarity and higher brightness.Such phenomena have led to a significant decline in the utilization of images,so the restoration of clear and haze-free images has certain research value and practical significance.Firstly,the paper analyzes the imaging principle of hazy images in depth,and takes the atmospheric scattering model as the research basis,and proposes the defects in the dark channel prior dehazing algorithm.Then,based on the Retinex theory,three fast and effective dehazing algorithms are proposed for these defects:(1)Aiming at the problem of incomplete haze removal and inaccurate transmission estimation for the dark channel prior algorithm,the image dehazing algorithm based on compensated dark channel and image fusion is proposed.Firstly,the algorithm compensates the dark channel with median filtering to solve the problem that the edge details of the restored image are not prominent;Then,the transmission is estimated simply according to the compensated dark channel,and based on the white balance theory,the transmission is estimated fuzzy;Finally,pixel-level image fusion is performed on the basis of the two methods,and the accurately estimated transmission is obtained.The experimental result shows that the algorithm can obtain more accurate transmission,eliminates the residual hazy phenomenon in the restored image,and improve the operation speed.(2)Aiming at the problem that the dark channel prior algorithm fails in the sky region,two adaptive parameters of the transmission estimation dehazing algorithm are proposed: 1)An adaptive multiscale transmission estimation dehazing algorithm.Firstly,the algorithm uses three Gaussian functions of different scales to respectively act on the RGB channels of the hazy image to obtain a “pseudo” dehazing image;Secondly,the adaptive parameter is obtained by using the mixed channel of hazy image,the parameter and the filtered minimum value are together acting on the “pseudo” dehazing image,and in order to obtain more accurate transmission,this paper uses joint bilateral filtering operation to eliminate the texture effect;Finally,this paper adopts the local-based atmospheric light estimation strategy and obtains the haze-free image based on physical models.The experimental result shows that the proposed algorithm has less running time.2)An adaptive linear transmission estimation dehazing algorithm.Firstly,the algorithm establishes a linear transformation model between the hazy image and the haze-free image minimum channel;Secondly,the adaptive parameter is obtained by using the mixed channel of the hazy image,and the transmission is estimated by combing the adaptive parameter and the linear transformation model;Next,constructing a Gaussian function through the minimum channel of the hazy image to compensate for the transmission in bright region,and improving the accuracy of the transmission of this region,and then use the cross bilateral filter to eliminate the texture effect to obtain accurate transmission;Finally,a clear haze-free image is restored with the atmospheric scattering model.The experimental results show that the algorithm can effectively reduce the time complexity,and the restored image has the advantages of obvious details and appropriate brightness.