A Kernel View of Spectral Point Pattern Matching
This paper investigates spectral approaches to the problem of point pattern matching. Specifically, kernel principle component analysis (kernel PCA) methods are studied and compared with Shapiro and Brady’s approach and multidimensional scaling methods on both synthetic data and real world data. We demonstrate that kernel methods can be effectively used for solving the point correspondence matching problem with a performance that is comparable with other iterative-based algorithms in the literature under the existing of outliers and random position jitter. We also provide discussion of the theoretical support from kernel PCA to the earlier approach of Shapiro and Brady.
KeywordsFeature Point Point Pattern Polynomial Kernel Kernel Principal Component Analysis Point Correspondence
- 1.Caelli, T., Kosinov, S.: An eigenspace projection clustering method for inexact graph matching. IEEE Tran. PAMI 26(4) (2004)Google Scholar
- 3.Carcassoni, M., Hancock, E.R.: Correspondence matching with modal clusters. IEEE Tran. PAMI 25(12) (2003)Google Scholar
- 5.Ham, J., Lee, D.D., Mika, S., Schölkopf, B.: A kernel view of the dimensionality reduction of manifolds. Max Planck Institute for Biological Cybernetics, Technical report TR-110 (2003)Google Scholar
- 6.Luo, B., Hancock, E.R.: Matching Point-sets using Procrustes alignment and the EM algorithm. BMVC (1999)Google Scholar
- 8.Scott, G.L., Longuet-Higgins, H.C.: An Algorithm for Associating the Features of Two Images. In: Proc. Royal Soc. London. Series B-Biological, vol. 244 (1991)Google Scholar