Recently, several image matchers based on Affine Invariant Regions have been developed and represent the state of the art for finding correspondences between two images taken from different viewpoints. This dissertation builds upon existing region detectors and develops various novel techniques which extend the power and functionalities of the regions.
The main contribution is a powerful Object Recognition system capable of working with very large scale and viewpoint changes, extensive clutter, and severe occlusion. It can handle non-rigid deformations, and also finds the approximative contours of the object.
The other innovations include a method to obtain correspondences across multiple images, a real-time affine tracker, and a technique to find coplanar region groups.
Vittorio Ferrari studied computer science with a minor in philosophy at the university of Fribourg, Switzerland, from which he graduated with a Dipl. Informatik in 2000. In the same year, he joined the Computer Vision Lab of the ETH Zurich, where he worked as a PhD student and research assistant. In 2004 he finished his doctoral thesis on affine invariant regions and object recognition, and was awarded a PhD degree (Dr. sc. tech.) from the ETH Zurich.

Keywords: object recognition, multiple views, wide-baseline stereo, visual tracking, augmented reality, texture segmentation