Novel eigenvector approach to pose and correspondence estimation

This paper proposes a novel eigenvector approach for pose and correspondence estimation between the feature points of two images or two point patterns under affine transformation. In the method, the proximity matrices, which record the normalized area features extracted from the two point sets are utilized to calculate the modes of each point set and the corresponding feature vectors. Then the point correspondence can be obtained by calculating the correlation of the feature vectors. To reduce the computation time, the idea of the principal component analysis (PCA) is adopted, which considers only the principal eigenvectors corresponding to the larger eigenvalues of each proximity matrix. As compared with the traditional eigenvector algorithm proposed by Shapiro and Brady, the proposed algorithm is demonstrated to be more effective in estimating the point correspondence and the relevant parameters of affine transformation.