Research On Single Image Dehazing Algorithm Based On Physical Imaging Model

Images are an important way for humans to obtain information.Humans can understand the world and make judgments by analyzing the images.With the rapid development of computer technology,images have been applied to various places such as public transportation safety,remote sensing satellite images,outdoor video surveillance,and applications in these places often require high-quality images to ensure the integrity and correctness of information.With the acceleration of China's urbanization construction process and the improvement of the national economic level,a series of factors such as the growing size of cities and the increasing exhaust emissions of vehicles have caused the haze problem in recent years.In haze weather conditions,suspended particles in the air increase,and the suspended particles in the air scatter and refract atmospheric light,which attenuates the intensity of light reaching the human eye or imaging equipment and causes problems such as blurred images,missing details,and incomplete information.In order to adapt the image acquisition equipment to the haze weather,the image dehazing technology came into being.This thesis researches and improves the problems of edge loss of detail,failure of atmospheric light estimation,and edge halo in the traditional dehazing algorithm based on physical imaging model.The research contents of this thesis are as follows:(1)After clarifying the research background and significance of the subject,this thesis studies and simulates the dark channel prior algorithm,learns the basic principles of the dark channel prior algorithm,and studies its shortcomings and defects.(2)Aiming at the problem that the over-bright area invalidates the overall atmospheric light estimation during atmospheric light estimation,this thesis improves it and introduces a simple linear iterative clustering method in image segmentation to select the ideal segmentation threshold to reduce the impact of bright areas on the overall atmospheric light estimate;Based on statistical results,choose an ideal segmentation threshold to reduce the influence of bright areas on the overall atmospheric light estimation value;And this thesis uses image fusion technology to fuse the luminance component(V space component)in HSV space and dark channel image to further accurately estimate the atmospheric light.(3)Aiming at the halo effect due to the abrupt change in transmittance in the scene's sudden change in depth,this thesis introduces the mean standard deviation method to determine whether it is a sudden change in the depth of field.And this thesis uses a weighted transmittance method to eliminate the halo effect in the scene's sudden change in depth and uses guided filtering for image restoration.(4)This thesis conducts simulation experiments.And a comparison is made between subjective vision and objective evaluation indicators with practical significance for image dehazing to verify the credibility of the experimental results.Experimental results show that the improvement of the transmittance estimation method in this thesis further eliminates the halo effect due to the inaccuracy of the transmittance in the sudden change of the scene depth;Through the method of fusing simple linear iterative clustering with images,the atmospheric light value estimation area is modified to reduce the influence of non-sky bright areas on the overall atmospheric light value estimation,so that a more accurate atmospheric light value is obtained.And the restored image has better results in terms of subjective vision,detail information,and visible edge contrast.