Improved SIFT Image Registration Algorithm Based On Multi-Scale Space

Excessive interference points when extracting feature points for SIFT(Scale-Invariant Feature Transform)registration algorithm,low registration accuracy and low registration efficiency.A two-layer SIFT registration algorithm based on scale space Marr wavelet is proposed.First,using a combination of scale space model and Marr wavelet,the reference image and the image to be registered are convolved at different scales,maximize the convolution result,the obtained maximum point is used as the feature point of the image.Then,feature matching is performed on the image using a two-layer SIFT registration algorithm.First layer coarse registration: feature matching of reference image and image to be registered using mutual Euclidean distance.The image is registered using the Euclidean distance from the reference image to the image to be registered,and the image is registered using the Euclidean distance from the image to be registered to the reference image.Leave the same matching point pair in the two registration results as a coarse registration result.The second layer is finely registered,and the results of the coarse registration are finely registered,and the model of the Random Sampling Consensus(RANSAC)algorithm is set.The results of the coarse registration are iterated according to the model,and the non-conforming point pairs are eliminated,leaving a match point pair that matches the model.To achieve accurate registration results.This paper will evaluate the performance of the algorithm by the registration of ordinary images and remote sensing images.It can be seen from the results that the registration accuracy of the algorithm for feature point matching of ordinary images is about 70% on average.The registration accuracy of feature point matching for remote sensing images is about 60% on average.Compared with the results of ordinary images and remote sensing images,the registration accuracy of the SIFT registration algorithm is improved by about 20%.Compared with the literature [62],the registration accuracy is improved by about 15%.At the same time,in the registration efficiency,the efficiency of the algorithm for registration of common images and remote sensing images is more than double the efficiency of the SIFT registration algorithm.It is verified that the registration accuracy and registration efficiency of the proposed algorithm are high,and the algorithm has certain feasibility and effectiveness.