SHAPE ENERGY MATCHING USING DEFORMABLE TEMPLATES

Abstract

Snakes, or) active contours, have been previously used in computer vision applications to locate and identify objects. However, problems associated with initialization, pdor convergence to boundary concavities and high computational I complexity, ha e limited their utility. In this thesis, we have proposed and

implemented a general non-occluded object localization and classification scheme I using deformable templates. Prior knowledge of an object shape is described by an I The shape variafions in an object class are achieved by combining a stable, invariant

and unique coniour model with Markov random field. The deformed shape contour I then interacts with the input image via a directional edge potential field calculated

from the salient) edge features. A Bayesian scheme, which is based on the prior knowledge and the edge information in the input image, is employed to find a match between the defoked template and objects in the image.

To avoid the high computational complexity, a coarse-to-fine algorithm was I implemented in n efficient hierarchical fashion. We have successfully applied the I suggested algorithm for the detection and classification of industrial parts. Results I show that the scheme is very robust with respect to scale, position and orientation I changes of the obj\ects as well as noise and local deformations of the shape.

I