Fast Method For Image Registration Based On SIFT Algorithm

In recent years,the technology of image registration has been widely used in the fields of architecture,medicine,military science and so on.In practical application,because it is difficult to get the overall appearance of the image,So we can get the local image and then use the image registration technology to splice the local image into the whole image of the desired image.The speed and accuracy of image registration is a hot and difficult problem in the field of image processing.It is of great significance to improve the speed of image registration and the quality of image mosaic.The image registration is a technology to combine two or more images with common parts into a complete image.There are three kinds of methods to implement the technology,they are respectively based on the image gray information,transform domain and image features.In this paper,the SIFT(Scale-invariant feature transform)algorithm is based on the image features of the image registration algorithm.This algorithm has a good effect on image invariant feature extraction,however,for some specific images,such as: noise,pixel height and some of the information in the location of the image,the operation speed of the algorithm is slow,and can not guarantee the specific requirements of high real-time.In this paper,the principle and process of SIFT algorithm and its deformation algorithm are discussed in detail,and comparative analysis of the experimental process,on this basis,some improvements are made in the application of SIFT algorithm to some specific image operations.This improvement is to use the SVD(Singular value decomposition)method to compress the original image,by compression,and select the appropriate number of eigenvalues,the feature point descriptor is reduced,thus reducing the computation of the algorithm,and achieve the purpose of reducing the computation time.By the analyze for the results of numerical experiments,it is proved that the number of feature values in the process of image compression is proportional to the total number of feature points extracted by the algorithm,and the total number of feature point descriptor is proportional to the average operation time of the algorithm,in other words,in this paper,the improved SIFT algorithm improves the speed of the algorithm.The improved algorithm ensures the accuracy of the original,while the computational speed is faster than the original algorithm,which provides a good guarantee for the work of image registration.