Design And Implementation Of Image Dehazing Algorithm

In recent years,our country's economic level has achieved rapid development,and the degree of industrialization has also been greatly improved.However,it has also caused some environmental problems,and haze pollution is one of them.The haze weather with low visibility has brought certain impacts on people's daily life,traffic supervision,and navigation.The images taken in the haze weather will be degraded to different degrees,and the details of the images will be lost,so the effect of identifying and processing degraded images in the later stage will be reduced.Therefore,in order to restore the lost details of the haze image,the study of efficient and practical image defogging algorithms has far-reaching research significance and application value.At present,the processing methods for haze images can be divided into two types:(1)defogging algorithm based on image enhancement;(2)defogging algorithm based on image restoration.This paper focuses on the defogging algorithm based on image restoration,using the atmospheric scattering model to restore the detailed features of the haze image.The main works are as follows:(1)This paper studies and analyzes the defects and shortcomings of the dark channel prior and color attenuation prior dehazing algorithm,I proposed a defogging algorithm based on weighted transmittance fusion.By constructing a weighting factor for transmittance,the defogging efficiency of the dark channel prior algorithm is combined with the high applicability of the color attenuation prior.Presents the advantages of the two algorithms,and introduces the brightness feature to make the transmittance have adaptive performance.(2)This paper proposes an image defogging method based on sky region segmentation.Through in-depth observation,research and analysis of the dark channel prior algorithm,I get the conclusion that the algorithm has an obvious defogging effect for most image areas.However,it is not applicable to the processing of areas with large brightness,such as large sky areas.The reason is that the atmospheric light value of the sky area is estimated incorrectly Therefore,in response to this shortcoming,this paper proposes sky region segmentation processing for haze images containing large sky regions,using the maximum between-class variance method(OTSU)and support vector machine(SVM)methods to analyze the sky.The region is segmented,the segmented images obtained are respectively defogged,and the results of the two segmentation methods are compared with the dark channel prior algorithm and the transmittance fusion algorithm.