Research On Low Quality Video Image Enhancement Algorithm For Dynamic Scene

In order to overcome the influence of low image quality caused by fog and rain weather,and the reduced camera imaging accuracy that leads to low follow-up processing effect or even inability to complete the work task,it is necessary to enhance the low-quality images captured in rain and fog weather.Aiming at the problem of image degradation in traffic scenes,this topic has carried out research on low-quality video image enhancement algorithms for dynamic scenes,overcoming the problems that traditional weather classification methods cannot classify images in any scene,the existing defogging algorithms have low image processing results that do not conform to assumptions,and the previous neglect of water vapor in rain and the complexity of real rain images.This topic mainly includes the following aspects:(1)Aiming at the problem that traditional weather classification methods cannot classify images in arbitrary scenes,a weather classification algorithm based on dual attention mechanism is proposed.Firstly,the channel attention module is proposed to extract weather features;Secondly,by combining the complementary characteristics of channel space dual attention,information is supplemented between different channels to improve the system recognition rate;Finally,a second-order feature network is added on the basis of dual attention.By reducing the maximum sample value feature and stretching the minimum sample value feature,the accuracy of the entire connection layer is further improved to achieve weather classification in any scene.The experimental results show that the accuracy of the weather classification method proposed in this paper is greatly improved compared with the existing methods,and has strong scalability and robustness.(2)Aiming at the problems that the existing defogging algorithms have low image processing results that are inconsistent with the assumptions and prior knowledge,and that the fog concentration in different regions is different in the real fog map,a defogging algorithm based on the fog line prior and attention fusion is proposed.First,the dark channel prior algorithm is used to obtain the initial transmission image,but because of local filtering,edge effect appears;Secondly,the transmission map is optimized by combining the fog line prior and feature selection attention fusion module,which not only overcomes the edge effect generated when the dark channel prior knowledge is used to obtain the transmission map,but also overcomes the appearance of fuzzy details and color distortion when the fog line prior knowledge is used separately;Finally,the quadtree hierarchical search method for atmospheric light value overcomes the color deviation phenomenon when the atmospheric light value is obtained interactively,and enhances the overall defogging effect.Experiments show that compared with other methods,this method has a better defogging effect,and the transition between frames is more natural;the edge details of the image are better preserved.(3)Aiming at the problems of neglecting rain in traditional rain removal work,which is often accompanied by fog or water vapor,and the complex background of the real rain map,etc.In this paper,the rain map model of rain fog mixture is constructed,and the research is carried out from the perspectives of rain drop detection and rain drop elimination.Firstly,we use the fluctuation difference between rain pixels and non-rain pixels and K-means clustering algorithm to judge the threshold value to achieve preliminary classification;Secondly,in order to avoid false detection and missed detection,combined with the color difference characteristics of rain pixels and background pixels,the optimized classified rain pixels and background pixels are obtained through secondary threshold judgment;Finally,in order to make up for the loss of background information in the process of rain map enhancement,the distilling block of detailed supplementary information is proposed to realize the adjustment of feature information in the way of gradual fine-tuning,so as to achieve the purpose of rain removal without damaging the background information.The test results show that this method has significant advantages over other methods in terms of subjective visual effect and objective evaluation.