Research On SAR Image Denoising And Multi-source Images Registration Algorithm

Synthetic Aperture Radar is an active remote sensing observation method with all-weather and all-time characteristics.These characteristics enable SAR to obtain effective information in complex environments.Therefore,SAR is particularly suitable for the acquisition of ground features in extreme weather areas,quick access to sudden disasters,monitoring of crustal deformation,etc..Since SAR images are heavily affected by speckle noise,interpretation of SAR images is very difficult.Meanwhile,since SAR image denoising is the basis of subsequent image processing,it is of great significance to study SAR image denoising.Image registration is the process of matching and superimposing two images obtained at different times,different viewing angles or under different conditions(climate,brightness,camera position and angle,etc.).With the rapid development of remote sensing technology and the emergence of many new sensors,people's ability to obtain multi-source remote sensing data is constantly improving.Multi-source registration is a key step in the field of image processing,and it is also the basis of multi-source image fusion and change detection.Therefore,the research on multi-source image registration is of great significance.This thesis focuses on SAR image denoising and multi-source image registration methods.The main contents are as follows:(1)The current SAR image denoising algorithms can achieve good smoothing effect,but they cannot preserve image details well.The research is conducted on this issue.Taking advantage of the characteristics of WNNM(Weighted Nuclear Norm Minimization)algorithm in restoring and maintaining image detail information,a SAR image denoising algorithm based on WNMM and FANS(Fast Adaptive Non-local SAR)is proposed.This algorithm improves the WNNM algorithm by using similarity measurement criteria suitable for SAR image speckle noise model to find similar blocks.Also,this algorithm takes advantage of the superiority of WNNM in image restoration and preservation and combines the good denoising performance of the three-dimensional wavelet coefficient shrinkage method in FANS.Therefore,this method improves the PSNR and SSIM of the SAR image and obtains a good denoising effect.(2)SIFT algorithm is sensitive to noise when calculating gradient at low octave.The research is conducted on this issue.Replacing the original gradient calculation with the ratio of the GGS(Gaussian-Gamma-Shaped)bi-windows and reconstructing the feature descriptor to improve the SIFT algorithm.At the same time,the dimension of the feature descriptor is expanded to 256 dimensions to improve the registration accuracy and stability of SAR images.Meanwhile,the algorithm can also achieve good registration results in multi-source images with small gray differences.(3)Currently,there is no widely applicable and stable multi-source image registration algorithm to deal with the problem of multi-source image registration with large grayscale differences and distortions.The research is conducted on this issue.The registration methods based on straight line features are studied,and the VSPM(Voronoi Spectral Point Matching)method is emphasized.However,it is found that VSPM method is very sensitive to external points.Therefore,under the framework of the method of coarse registration and fine registration,a new method combining straight line and virtual intersection of straight lines is proposed.This method can still be applied when there are large grayscale differences and distortions between multi-source images,and can realize the registration of multi-source images with obvious straight line features and at least three sets of non-parallel straight line segments in the image.