| In modern life,intelligent monitoring systems,unmanned driving technology and other fields have made great progress.Among them,computer vision systems have a wide range of applications and indispensable positions in this field.When atmospheric phenomena such as fog and haze occur,the presence of suspended particles in the air will reflect and refract the reflected light and atmospheric light of the object,the normal operation of the computer vision system will be affected,the external environmental information collected by the computer vision system will be missed,and the difficulty of extracting real and useful information from environmental images will be increased.Therefore,improving the image clarity in fog and haze weather and restoring part of the lost information in the image can improve the effectiveness and reliability of the operation of the computer vision system,which is of certain practical significance.At present,most of the application scenarios of the image dehazing algorithm are outdoor daytime,but the night scene mostly contains artificial light sources that do not exist in the daytime scene,the fogging effect of the daytime scene dehazing algorithm is reduced or even invalid in the nighttime scene.In this paper,various dehazing algorithms for daytime and nighttime scenes are analyzed,and two dehazing algorithms are proposed,which are respectively aimed at haze-containing images of daytime scenes and nighttime scenes.The specific research content is as follows:(1)Aiming at the problems of sky distortion,edge artifacts and overall darkening of restored images in existing dehazing algorithms,a daytime image dehazing algorithm based on linear transformation and HSV color space is proposed.Firstly,the fog image is preprocessed,and the highlighted area of the image is compressed by nonlinear adaptive compression.Secondly,in the phase of media transmission rate calculation,the compressed image containing fog is converted into HSV color space,and the linear transformation model of media transmission rate calculation is established by combining it with the linear transformation theory.Finally,the atmospheric light value is obtained by the scoring strategy of the quadtree method optimized by position information.The direct purposes of the three improvements are to reduce the influence of the highlighted pixel in the image containing fog,improve the accuracy of medium transmission rate estimation and enhance the rationality of atmospheric light value selection.(2)Aiming at the problems of edge artifacts,overall darkness and color distortion in the light source region in the restored images of the mainstream night dehazing algorithm,a night image dehazing algorithm based on light source segmentation and dark and bright channel priors is proposed.Firstly,the color channel difference map of the fog image combined with a Gaussian pyramid is used to segment the light source.Secondly,in the phase of medium transmission rate calculation,the optimal fusion model of medium transmission rate is established through the initial medium transmission rate and the probability diagram of the light source region obtained by dark channel prior and bright channel prior.Finally,the probability map of the light source region is used as the guide map of guided filtering of images containing fog,and the atmospheric light value is obtained.The direct purpose of the three improvements is to improve the accuracy of light source segmentation,improve the accuracy of medium transmission rate and improve the expression ability of edge information in atmospheric light value.The experimental results verify that the daytime image dehazing algorithm based on linear transformation and HSV color space is more accurate in obtaining atmospheric light value and medium transmission rate,the time complexity is low,and the sky area of the restored image has natural color,moderate overall brightness,and good overall effect.Based on the light source segmentation and the dark light channel a priori the night image dehazing algorithm is more accurate for the segmentation of the light source area,on this basis,the more accurate medium transmission rate is obtained,the atmospheric light value contains more edge information,the overall color of the restored image is rich,the details and texture information are better restored and expressed,the overall dehazing effect is better,but the time complexity is slightly higher. |
