ELSEVIER Pattern Recognition Letters 18 (1997) 49- 53
Pattern, R.ecognition
Lerters
Shape alignment - Optimal initial point and pose estimation
Jorge S. Marques a,*, Arnaldo J. Abrantes b
a INESC, Technical University of Lisbon, R. Alves Redol 9, P-IO00 Lisboa, Portugal
b INESC, Polytechnic Institute of Lisbon, Lisboa, Portugal
Received 30 November 1995; revised 20 October 1996
Abstract
This paper describes an algorithm for the optimal alignment of a pair of 2D shapes defined by point sequences. The
proposed algorithm provides closed-form expressions for the estimation of the initial point, scale and pose parameters.
© 1997 Elsevier Science B.V.
Keywords: Shape alignment; Object representation; Pose estimation; Image analysis
1. Introduction
2D shape analysis plays a key role in many
computer vision systems. Several models were pro-
posed in the literature to represent the object bound-
ary, including parametric models, active contours,
differential equations and Fourier descriptors (e.g.,
Otterloo, 1991; Kass et al., 1987). One of the sim-
plest ways to represent an object shape consists of
forming a sequence of N boundary points, detected
in the image according to some criterion (e.g., He
and Kundu, 1991).
Let x denote the sequence of detected points and
a reference sequence adopted as a model. Several
difficulties arise if one tries to compare both se-
quences due to changes of the object attitude, scale,
shape and initial point.
Some of these difficulties have been extensively
studied. The estimation of the object pose and scale
from two sets of matched points is a well known
problem which has been addressed by many re-
* Corresponding author. E-mail: jsm@inesc.pt.
searchers and which can be efficiently solved, e.g.,
by robust least squares techniques (e.g., Haralick et
al., 1989). Shape deformation has also been ad-
dressed by several authors, especially in the context
of active contours (e.g., Kass et al., 1987). The
estimation of the initial point is however a less
studied problem. This task is often performed by
complex optimization procedures followed in some
cases by a manual correction.
This paper describes a new algorithm for the
optimal estimation of pose, scale and initial point
parameters. The proposed algorithm is computation-
ally efficient and allows a simple implementation.
Experimental tests show that the proposed technique
is able to align complex shapes and it is robust
against shape deformations and changes of the initial
point.
2. Problem formulation
Let x i ~ E2, i = 0 ..... N- 1, be a sequence of
N contour points detected in the image and orga-
0167-8655/97/$17.00 © 1997 Elsevier Science B.V. All rights reserved.
PH S01 67-8655(96)00 120- I