Method Research Of Fog Removal And Enhancement For Remote Sensing Images

Aiming at the problems of low saturation,color distortion and blurred details of remote sensing image caused by bad weather conditions such as cloud,fog and haze.Combining two typical image processing methods,image enhancement and image restoration,a remote sensing image defogging and enhancement algorithm based on improved dark channel prior theory and single-scale Retinex in HSI color space is proposed.Firstly,based on the image restoration method and the characteristics of foggy remote sensing image,in order to apply the general dark channel prior theory to foggy remote sensing image,the maximum value of pixels within the dark channel threshold range of foggy remote sensing image is selected as the estimated value of atmospheric light A0.In addition,the empirical parameter Gin Pan[46]algorithm is introduced to improve the general dark channel prior theory algorithm,and then the optimized transmittance is obtained.The experimental results show that the transmittance of the optimized image is improved,the edge information of the image remains intact,and the building and road information is clearer.Although the improved dark channel algorithm improves the problem of dim brightness and blurred edge details after defogging based on the general dark channel prior theory.However,there is still the problem of contrast reduction in varying degrees.Therefore,it is necessary to use the single scale Retinex theory defogging algorithm based on HSI color space to further enhance the color and detail of the restored image.Secondly,the color space of the restored image is converted from RGB space to HSI space.The hue component H is not changed,and the other two components are processed separately:(1)NSCT transform is used to decompose the brightness component I to obtain high-frequency information IH and low-frequency information IL.The high-frequency information IH is processed by guided filter edge preserving denoising,and the low-frequency information IL is processed by single-scale Retinex algorithm based on image enhancement method to enhance the edge details.The high-frequency information IH and low-frequency information IL were recombined and I’ was obtained by NSCT inverse transform.(2)The color linear stretching method is used to enhance the contrast of saturation component s,and S’ is obtained.ce of the restored image is converted from RGB space to his space.Finally,the reconstructed hs’i’space is reversed back to RGB space to get the remote sensing image after defogging.The combination of the two methods not only effectively improves the problems existing in He[10]algorithm and Pan[46]algorithm,but also solves the problem of color distortion of fog remote sensing image directly using single-scale Retinex algorithm.In this paper,the fusion algorithm improves the contrast and clarity of the image,so as to achieve the effect of defogging and enhancement.After defogging pretreatment,the fog remote sensing image can be fully utilized,which not only avoids the waste of resources,but also reduces the impact of bad weather conditions on remote sensing image.It is conducive to the subsequent visual interpretation,spectral analysis,recognition and classification,feature extraction and mapping output of the image.Taking the remote sensing image with medium spatial resolution as the experimental data source,the feasibility and effectiveness of the fusion algorithm are verified by subjective qualitative analysis and objective quantitative evaluation(such as information entropy,average gradient,peak signal-to-noise ratio and mean square error).