Abstract:

Registration based on SIFT feature points is a frequently used method. However, in complex background, SIFT feature points are usually largeamount and redundant, which brings in a number of problems such as a waste of storage space, mismatching and timeconsuming. In order to solve these problems, we propose a new SIFT feature extraction algorithm which can eliminate redundancy. The algorithm first extracts the SIFT feature points, and determines whether a feature point drops in the target area or not according to the gradient state around it. Then the feature points in the target area are retained while the feature points in the background area are deleted, in which way the number of feature points is reduced and the redundancy is also eliminated. The quality of feature points is finally judged by the ratio of points in the target area to those in the background area. Experiment results show that the proposed algorithm eliminates a large number of interference feature points, which improves the accuracy and efficiency of subsequent registration.

Key words: SIFT feature points;redundancy elimination;gradient;target area;background area