Key point reduction in SIFT descriptor used by subtractive clustering

Abstract

The SIFT descriptor is one of the most widely used descriptors and is very stable in regard to changes in rotation, scale, affine, illumination, etc. This method is based on key points extracted from the image. If there are many such points, a lot of time will be needed in the matching and recognition phases. For this reason, we have tried in this article to use the clustering technique in order to reduce the number of key points by omitting similar points. In other words, subtractive clustering is used to select key points which are more distinct from and less similar to other points. In the section on conclusions, a successful implementation of this method is presented. The efficiencies of the proposed algorithm and of the base SIFT algorithm on the data set ALOI were investigated and it was observed that by adding this method to the base SIFT descriptor the rate of recognition increases by two percent and the time complexity decreases by 1.035728 seconds.