Service Communautaire d'Information sur la Recherche et le Développement - CORDIS


INDET Résumé de rapport

Project ID: G4RD-CT-2002-00830
Financé au titre de: FP5-GROWTH
Pays: United Kingdom

Automatic image registration

Methodologies and algorithms have been developed to achieve automatic image registration based on ellipses, elliptic arcs, and non-uniform rational B-splines (NURBS). For image registration based on ellipses, an analytical solution has been developed for coarse matching and an objective function has been developed based on the non-overlapping area for fine matching.

For image registration based on elliptic arcs, a novel objective function has been developed based on integral of squared distances (ISD) between two elliptic arcs, and a hypothesis-verification type of search algorithm has been implemented for correspondence search.

For automatic image registration based on NURBS, a novel method has been developed which consists of rough 3D/2D alignment by estimation of the two 3D rotation parameters via comparison of the similarities between the silhouette extracted from the date image and the silhouette patterns projected using different viewing points; rough pose estimation by 2D similarity alignment of the image silhouette and the most similar projected silhouette; and pose refinement by iterative alignment of the model and the input data image based on the closest point method until the resulting error is sufficiently small.

These automatic image registration methods can be applied to match an image to its corresponding CAD model, or to match two images acquired from the same component at different times.


Lik-Kwan SHARK, (Head of ADSIP Research Centre)
Tél.: +44-1772-893253
Fax: +44-1772-892915