Research On Registration And Fusion Algorithm Of Remote Sensing Image Based On SIFT And Wavelet Transform

Remote sensing image data are continuously sent back to the ground receiving station.In face of such massive remote sensing images,image registration and fusion technology play a key role.The registration of remote sensing images is a necessary prerequisite for effective data fusion.The fusion of remote sensing images is to improve the availability of data by integrating complementary remote sensing images.The fused remote sensing images contain more information than the source images,thus making preparations for subsequent processing.This paper focuses on the registration of remote sensing images and the fusion of multispectral and panchromatic images,proposes corresponding improvement methods for the problems in the registration and fusion process,and verifies,evaluates and analyzes them.The main work of this paper is divided into the following three aspects:(1)The theories and methods of the commonly used remote sensing image registration and fusion algorithms are studied.The remote sensing image registration and fusion algorithms are compared,and the advantages and disadvantages of each algorithm are objectively analyzed.Compared with other registration algorithms,SIFT registration algorithm has obvious advantages.It can detect stable feature points for registration under different scales,rotation,illumination and other conditions.Its disadvantages are long operation time and edge effect.Compared with other fusion algorithms,the wavelet transform fusion algorithm can carry out multi-scale fusion,so as to improve the details of the fused image.Its disadvantages include block,spectral distortion,edge blur and so on.(2)An improved image registration algorithm for SIFT remote sensing image based on Canny and PCA was proposed.Canny algorithm and SIFT algorithm were used to detect the edge features of the reference image and the image to be registered respectively.The edge effect was reduced by eliminating the feature points on the edge features,and then the dimensionality was reduced by using the feature point descriptors of PCA.Finally,the method of RANSAC removing mismatched pairs was used to complete the registration.Thus,a large number of edge response points are removed to improve the efficiency of the algorithm.(3)A multispectral and panchromatic remote sensing image fusion algorithm based on PCNN wavelet transform is proposed.IHS transform and PCA transform were applied to the multispectral images.After IHS transformation,intensity I,chromaticity H and saturation S are obtained,and PCA was transformed to the first principal component PC1 of the multispectral image.The first principal component PC1 will panchromatic image histogram matching with the first principal component PC1 to PAN.Wavelet transform was applied to I and PAN' to obtain the low frequency component and the high frequency component.Weighted average fusion rules were used for the low-frequency components,and improved PCNN fusion rules were used for the high-frequency components,and then I' was generated.Finally,the fusion results were generated by IHS inverse transformation.This algorithm effectively reduces the spectral distortion and improves the fusion details.