Research On Technology And Application Of Defogging Based On Haze Weather Image

Under the influence of haze weather,the image acquired by optical imaging equipment will degrade,and the degraded image with fog seriously reduces the application value of the image in various fields.Image defogging technology can reduce or eliminate the negative impact of haze on the image,so it becomes an important research direction in the field of image processing.In this paper,the problems existing in the existing image de-fogging algorithm are studied.The main work is as follows:1.Aiming at problems such as the failure of dark channel prior algorithm in some areas,inaccurate estimation of atmospheric light value and dark color of de-fogging image,this paper proposes an improved dark channel prior image de-fogging algorithm.Firstly,in order to improve the accuracy of the calculation of atmospheric light value,an improved quadtree segmentation algorithm is proposed,which can effectively avoid the interference of the location of atmospheric light region in the scene of highlighted region and bright ground.Secondly,the light channel priors are used to make up for the failure part of the dark channel priors.The atmospheric light value is taken as the threshold to obtain the transmittance of the light and dark regions respectively,and the comprehensive transmittance of the image is obtained.The coarse transmittance is modified and refined by guided filtering.Thirdly,the color correction algorithm is introduced to enhance the brightness of the image,so that the image is more in line with people’s visual feelings.Finally,the comparative experimental results show that the proposed algorithm has better restoration effect.2.Aiming at the problems of parameter error amplification caused by parameter estimation of current image de-fogging algorithms based on deep learning and high complexity of existing algorithms,an integrated network image de-fogging algorithm based on depth-separable convolution is proposed in this paper.Firstly,deep separable convolution is used to replace the conventional convolution of the original algorithm,reducing the number of network parameters and making the network more lightweight.Secondly,aiming at the problem that the original network directly connected the feature maps without highlighting the key feature information,a weighted feature fusion module was designed to assign different weights to the feature maps of different levels and then weighted fusion.Thirdly,Multi-Scale Structure Similarity(MS-SSIM)and mean square error weighting are used as loss functions to optimize the contrast,brightness and color saturation of the de-fogging image,so that the restored image is more in line with people’s visual perception.Finally,the comparative experimental results show that the proposed algorithm not only ensures the lightweight of the network model,but also enhances the de-fogging effect to some extent.3.Combined with the algorithm of de-fogging proposed above,this paper designs a license plate recognition system in hazy weather.First of all,the license plate recognition of the real application scenario design,combined with specific scenarios to do demand analysis of the system,to identify the problems to be solved.Secondly,aiming at the problem that the vehicle license plate information in the haze weather is difficult to recognize in the existing applications,the license plate recognition system has carried out detailed module design and function design.Finally,a comparative experiment is carried out,and the experimental results show that the license plate recognition system proposed in this paper can effectively recognize vehicle license plate information in haze weather,and has certain application value.