SIFT with the F-transform pre-processing

Abstract

This paper analyzes the efficiency of the Scale Invariant Feature Transform (SIFT) measured in terms of execution time and stability of detection for keypoint location. Our goal is to investigate the influence of various downscaling methods which pre-process images before the SIFT is executed. For every downscaling method, we estimate the total execution time, the amount of detected keypoints and the success rate of keypoint matching. We show that the best combination with the SIFT is achieved by the F-transform downscaling algorithm.