Gaofen No. 2 Panchromatic And Multispectral Remote Sensing Image Registration Plug-in Research And Development

GF-2 satellite can provide panchromatic remote sensing images with spatial resolution better than 1m and multispectral remote sensing images with spatial resolution better than 4m,and the registration efficiency of panchromatic remote sensing images and multispectral remote sensing images will directly affect the results of the following image processing and analysis.When the image registration method based on SIFT is used to register panchromatic and multispectral remote sensing images,the registration speed is slow and the registration accuracy is low.To solve this problem,an improved SIFT remote sensing image registration algorithm based on information entropy grid partitioning is proposed in this thesis,and the improved algorithm is integrates into the remote sensing information processing platform MHMap GIS in the form of plug-ins.The main work of this thesis is as follows:Firstly,in view of the fact that SIFT algorithm has a large number of feature points and needs to take a long time to be extracted,this thesis adopts the strategy of information entropy grid division,and only detects the feature point in areas with a large information entropy.So the calculation amount of feature extraction and the number of feature points can be reduced and simplified.Secondly,in the procedure of generating scale space by SIFT algorithm,the basal image is downsampled without interpolation,which lead to inaccuracy problems of the feature points when rotation is involved.So Bilinear interpolation in this thesis is used to replace the original SIFT algorithm in order to improve the accuracy of image registration indirectly.And then,to solve the problem that the feature points extracted by SIFT algorithm contain many unstable edge response points,Canny operator is introduced to detect the edge to obtain the edge coordinate points.Then the coordinates of feature points detected by Canny operator and SIFT algorithm were compared to further eliminate the unstable edge response points.Lastly,in order to solve the problem that SIFT algorithm has a low accuracy in the registration of remote sensing image registration,this thesis gives a new gradient calculation method.What's more,this thesis also improved the SIFT feature descriptor generation method.That is,in the feature matching stage,the mutual Euclide distance between the reference image and the image to be registered is taken as the constraint condition of the initial matching,which combine the FSC algorithm to realize the fine matching.Finally,according to the system integration requirements provided by MHMap GIS,a plug-in interface program is written to integrate the algorithm in this thesis into the software platform in the form of a plug-in.In this thesis,registration experiments are carried out on the GF-2 remote sensing image pairs of different regions and different phases,and the improved algorithm and the comparison algorithm are analyzed and compared.The experimental results show that the algorithm in this thesis can obtain more correct matching point pairs on the premise of obtaining fewer feature points.And it is also better than the comparison algorithm in terms of registration speed and registration accuracy,so it can achieve the sub-pixel registration accuracy and meet the project requirements.What's more,after the integration of algorithm plug-in,the overall operation effect is good and meets the requirements of system integration.