Image Registration Method Based on Optimized SURF Algorithm

Zhang Sheng,Zhou Meng,Qian Mingsi,Li Peihua,Liu Yuli,Ji Changgang

doi:10.11648/j.ajop.20190704.11

Abstract

In order to solve the time consuming problem of image registration based on the traditional SURF algorithm, the image registration method based on the optimized SURF algorithm is proposed. Firstly, the image corner points are extracted by the Shi-Tomasi algorithm, then, the SURF algorithm is used to generate the corner point descriptors and the sparse principle algorithm is used to reduce the dimension of the corner point descriptors. Finally, the bidirectional matching algorithm is used to match. Through the experimental data analysis, the image registration method based on the optimized SURF algorithm is nearly the same in image registration accuracy in comparison with the traditional SIFT algorithm, the traditional SURF algorithm and the other four optimized algorithms, but the time consuming of image registration is decreased by 79.09%, 47.74%, 66.25%, 50.79%, 21.43% and 5.13%, respectively, verifying the instantaneity and effectiveness of the algorithm.

Highlights

Image registration methods are widely used in the fields of medical image processing, remote sensing image processing and panoramic stitching [1,2,3]
We use the traditional SIFT algorithm [21, 22], the traditional SURF algorithm, the reference [16] algorithm, the reference [17] algorithm, the reference [18] algorithm, the reference [19] algorithm and the image registration method based on the optimized SURF algorithm to carry out the image registration experiment on the experimental registration images
According to figure 4 (b), compared with the traditional SIFT algorithm, the traditional SURF algorithm, the reference [16] algorithm, the reference [17] algorithm, the reference [18] algorithm and the reference [19] algorithm, the image registration accuracy of the image registration method based on the optimized SURF algorithm is equivalent

Summary

Introduction

Image registration methods are widely used in the fields of medical image processing, remote sensing image processing and panoramic stitching [1,2,3]. In order to solve the time consuming problem of image registration based on the traditional SURF algorithm, many optimization algorithms had been proposed by domestic surgical researchers in recent years. Aiming at the time consuming problem of image features extraction based on the traditional SURF algorithm. Aiming at the time consuming problem of image features matching based on the traditional SURF algorithm. In 2014, Yang Sa et al proposed to use the SRP (sparse random projection) algorithm to reduce the dimension of the image feature descriptors, so as to reduce the time of the image features matching [9]. In order to solve the time consuming problem of image registration based on the traditional SURF algorithm, this paper uses the Shi-Tomasi algorithm [14], the sparse principle algorithm [15] and the bidirectional matching algorithm to optimize the traditional SURF algorithm. The result is that the accuracy of image registration is equivalent and shorten the time of image registration, namely the Shi-Tomasi-CS-SURF algorithm

Shi-Tomasi-CS-SURF Algorithm

Corner Extraction

Corner Point Description and Sparse Principle

Bidirectional Matching

Experimental Results

Conclusions