Ordinal-Measure Based Shape Correspondence
- PDF / 1,890,632 Bytes
- 10 Pages / 600 x 792 pts Page_size
- 58 Downloads / 217 Views
Ordinal-Measure Based Shape Correspondence Faouzi Alaya Cheikh Signal Processing Laboratory, Tampere University of Technology, P.O. Box 553, FIN-33101 Tampere, Finland Email: [email protected]
Bogdan Cramariuc Signal Processing Laboratory, Tampere University of Technology, P.O. Box 553, FIN-33101 Tampere, Finland Email: [email protected]
Mari Partio Signal Processing Laboratory, Tampere University of Technology, P.O. Box 553, FIN-33101 Tampere, Finland Email: [email protected]
Pasi Reijonen Signal Processing Laboratory, Tampere University of Technology, P.O. Box 553, FIN-33101 Tampere, Finland Email: [email protected]
Moncef Gabbouj Signal Processing Laboratory, Tampere University of Technology, P.O. Box 553, FIN-33101 Tampere, Finland Email: [email protected] Received 31 July 2001 and in revised form 10 February 2002 We present a novel approach to shape similarity estimation based on distance transformation and ordinal correlation. The proposed method operates in three steps: object alignment, contour to multilevel image transformation, and similarity evaluation. This approach is suitable for use in shape classification, content-based image retrieval, and performance evaluation of segmentation algorithms. The two latter applications are addressed in this paper. Simulation results show that in both applications our proposed measure performs quite well in quantifying shape similarity. The scores obtained using this technique reflect well the correspondence between object contours as humans perceive it. Keywords and phrases: shape, ordinal, correlation, content, retrieval, indexing, segmentation, performance.
1.
INTRODUCTION
Shape representation techniques are generally characterized as being boundary-based or region-based. The former (also known as contour-based) represents the shape by its outline, while the latter considers the shape as being formed of a set of two-dimensional regions. The human visual system itself focuses on edges and ignores uniform regions [1, 2]. This capability is hardwired into the retina. Connected directly to the rods and cones of the retina are two layers of the neurons that perform an operation similar to the Laplacian. This operation is called lateral inhibition and helps us to extract boundaries and edges. Therefore, in this paper we focus on this aspect of the shapes and not on the regions they may contain. Object contours however, will have intrinsic intraclass variations. Moreover, object boundary deformation is expected in most imaging applications due to the varying
imaging conditions, sensor noise, occlusion, and imperfect segmentation. Estimating the similarity between objects shapes can be described in a simplistic way in two steps: shape features extraction and feature comparison. Each of these two steps however represents a diļ¬cult problem by itself. Selecting a set of features to characterize a shape for a certain application is not easy, since one must take into consideration the variability of the shapes and the application domain specificity. Feature comparison
Data Loading...
