A novel approach of fingerprint recognition based on multilinear ICA

Motivated by the reported out performance in the fingerprint recognition thesises of PCA by ICA in the linear case where only a single factor is allowed to vary, and the outperformance of PCA by FET when multiple factors are allowed to vary, it is natural to ask whether there a multilinear generalization of ICA and if its performance is better than the other two methods. In this paper, We have presented a multilinear algebraic framework for fingerprint image recognition, which employs a tensor (N-mode) extension of the conventional matrix SVD. We also introduced a multilinear projection algorithm for fingerprint recognition, which projects an unlabeled test image into the N constituent mode spaces to infer its mode label-sperson, finger, type.