This project presents a comparative study of two competing algorithms for the task of detecting keypoints in images. One is the Harris corner detector and the other is the Scale-invariant keypoints approach (referred to as Keypoints approach). Also in this project the applications of "distinctive" image features extracted from the neighborhood of these keypoints/corners is demonstrated. A performance index called the matching ratio is proposed by which we can evaluate our naïve object recognition / image matching algorithms. Finally, the superior of the two algorithms is compared with the pruning based template approach in the IARC symbol recognition project undertaken by the UA Aerial Robotics Club.
Warning: My version of SIFT code does not include the interolation step of the actual algorithm. Also the initialization step is different. The results are somewhat deteriorated as compared to the actual SIFT algorithm (Lowe et al.).