Research On Optical And SAR Remote Sensing Image Registration Algorithm Based On Phase Consistency And SIFT

Image matching is an important technology in computer vision.The task of image matching is to transform two or more images into the same coordinate system,so as to make full use of various information.Nowadays,image matching plays an important role in artificial intelligence robot,weapon guidance,unmanned driving,three-dimensional reconstruction,medical image detection,text,fingerprint and face recognition.Remote sensing image matching also plays a very important role in map updating,disaster prevention and relief.But in the actual application process,there are many problems,such as different imaging principles,different resolutions,different shooting angles,different light contrast,image rotation,distortion and so on.Both optics and SAR images have their own advantages.Comprehensive utilization can complement each other’s advantages.However,there are some problems in the automatic matching of Optics and SAR images which are different from the imaging principle.First,the radiation characteristics of the two objects are very different,which will cause different gray-scale characteristics of the same name features in the two images.Second,the speckle noise of SAR image will also brings some difficulties to matching.This paper studies the image matching problem and improves the SIFT algorithm based on the phase consistency technology.The main research contents and results are as follows:1.In view of the problem that it is difficult to extract features of the same name from heterogeneous images,this paper uses the phase consistency technology to make up,generates a phase consistent image through the Log Gabor filter,and calculates the maximum distance map,and performs the feature point on the maximum moment image Extraction can achieve better results.2.It is proved and analyzed through experiments that neither the gradient map nor the maximum distance map is suitable for establishing feature descriptors,and then a ratio maximum index map(RMIM,Ratio Maximum Index Map)is proposed.Based on this,descriptor construction is carried out in order to To ensure that the algorithm has rotation invariance,a directional proportional maximum index map is proposed,and the ratio map and the proportional maximum index map are generated again after imitating the SIFT algorithm to rotate the window to facilitate the construction of feature descriptors.3.Use the Euclidean distance between feature vectors to measure their similarity,and use the FSC(Fast Sample Consensus)algorithm to eliminate outliers,and improve the correct rate through two-way matching.Experiments show that the proposed method based on phase consistency and SIFT can achieve the matching of optical and SAR images,and can also be applied to some other matching tasks.