A robust multi-constrained model for fingerprint orientation field construction

This paper proposes a robust multi-constrained model for fingerprint orientation field construction. The orientation field construction problem is formulated as an overdetermined regularization system, in which three constraints are incorporated to ensure the accuracy. Constraints required in the model are: 1) a least square data term aiming to maintain consistency between the constructed and original orientation field; 2) a total variation regularization aiming to smooth the orientation field and eliminate the global noise; and 3) a nuclear norm regularization aiming to eliminate sparse noise while preserve the structure of the orientation field. According to experiments on both high-quality fingerprint image and low-quality fingerprint image, the proposed model shows high performance in achieving accurate orientation field. Thanks to the short running time, the proposed model is applicable to applications such as fingerprint indexing.