Performance Evaluation of Scale Invariant Feature Transform

The SIFT algorithm produces keypoint descriptors. This paper analyzes that the SIFT algorithm generates the number of keypoints when we increase a parameter (number of sublevels per octave). SIFT has a good hit rate for this analysis. The algorithm was tested over a specific data set, and the experiments were conducted to increase the performance of SIFT in terms of accuracy and efficiency so as to provide its use in many real time applications for image recognition. We also shown that a keypoint is an image feature which is so distinct even though the image has gone through transformations i.e. given a keypoints in an image, if one scales the image to half the size, double the size, rotated to a particular degree, or object occlusion is done, the image can be recognized because some keypoints would still be identifiable. The algorithm is specifically tested for its feasibility for finding keypoint matches between two images.