Sparse coding based orientation estimation for latent fingerprints

Fingerprint orientations are often used to describe the ridge flow patterns, providing useful features for further fingerprint processing and recognition. Although significant advances have been achieved for orientation estimation, it is still challenging to reliably estimate 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 coding in image denoising and reconstruction, this paper proposes an orientation estimation algorithm based on dictionary learning and sparse coding for latent fingerprints. First, a texture image is obtained by decomposition of latent image with a total variation model. The structured noise is greatly reduced from the texture image. Second, we propose a multi-scale sparse coding method for iterative estimation of local ridge orientations on the texture image. Multi-scale dictionaries are learned from the orientation fields of good quality fingerprints to capture the prior knowledge of various orientation patterns, and sparse coding is iteratively applied with the increase of patch sizes to correct the corrupted orientations of latent fingerprint. The proposed algorithm can work well to reduce the effect of various noise and restore the corrupted orientations while maintain the details of singular region. Experimental results and comparisons on NIST SD27 latent fingerprint database are presented to show the effectiveness of the proposed algorithm.