On the Uniqueness of Curvature Features

Cicero Mota, Erhardt Barth

Institute for Signal Processing
University of Luebeck
Ratzeburger Allee 160, 23538 Luebeck, Germany
barth@inb.uni-luebeck.de
 

Abstract:

Curved image features like corners are good features to track and also the only image regions where motion can be estimated reliably. In addition, such features contain higher amounts of information than straight edges and uniform regions. Here we extend previous result related to the uniqueness of curvature features and proof that curved image regions contain all the information in the image. The proof is constructive and we present first reconstruction results. Finally, an extension of the proof for the case of image sequences and 3D is outlined.

Full paper in PDF format