Latent fingerprint orientation estimation via sparse representation

The local ridge orientations are often used to describe fingerprint ridge flow patterns, providing useful information for fingerprint recognition. Although significant advances have been achieved for orientation estimation, it is still challenging to reliably compute the orientations for latent fingerprints, which are usually of poor quality with unclear ridge structure and various overlapping patterns. Motivated by the recent success of sparse representation in image denoising and reconstruction, this paper proposes an orientation estimation algorithm based on dictionary learning and sparse representation for latent fingerprints. The proposed algorithm consists of two main stages which are offline dictionary learning and online orientation estimation. In the offline stage, to capture the prior knowledge of various orientation patterns, a redundant dictionary is learned from the orientation fields of good quality fingerprints. In the online stage, the total variation model is first applied to decompose the latent fingerprint into cartoon and texture components. The cartoon component with most of non-fingerprint patterns is removed as the noise, while the texture component consisting of the weak fingerprint is used for orientation estimation. Second, the initial orientation field is estimated with the traditional method and sparse representation on the redundant dictionary is employed to reconstruct the orientation field of latent fingerprint. Experimental results and comparisons on NIST SD27 latent fingerprint database are presented to show the effectiveness of the proposed algorithm.