Symmetrical null space LDA for face and ear recognition

摘要

Many natural objects such as face and ear manifest symmetry. The mirror images of symmetrical objects also encode significant discriminative information, which is of benefit to recognition performance. In this paper, a novel symmetrical null space method with the even-odd decomposition principle is proposed for face and ear recognition. By introducing mirror images, the two orthogonal even/odd eigenspaces are constructed. Then the discriminative features are, respectively, extracted from the two eigenspaces under the most suitable situation of the null space. Finally, all the features are combined for classification. Experimental results on both face database and ear database demonstrate the performance of the proposed method.