Evaluating the Robustness of Local Image Features: Object Recognition with Stable Features

In this thesis we examine a feature refinement technique that is based on scaling and rotating images prior to the features extraction. The main idea is that stable features of an image are not affected by any transformations of the image. As a consequence simulating these transformations and reducing the number of keypoints to the number of those keypoints that survive these simulations, results in a set of only stable keypoints for the image.