Affine invariant descriptors using principal components analysis
Ahmed El Oirrak · Pattern Recognition and Image Analysis · 2008
In this paper, we propose new methods for recognizing 2D/3D objects undergoing affine transforms. Robustness with respect to level-of-detail is achieved by selection of points belonging to fixed directions on a circle, i.e., the 2D case, or a sphere, i.e., the 3D case (called the Ray casting selection method in the literature). The proposed descriptors are based on principal component analysis (PCA); each shape is represented by its eigenvalues and the corresponding eigenvectors. The proposed methods allow recognition under an affine transform which is not possible using other methods in the literature, for example, that in [1]. Here we use an asymmetric PCA to achieve invariance under an affine transform.