PDE-based regularization of orientation field for low-quality fingerprint images

Orientation field is one of the intrinsic characteristics of fingerprints, and it plays very important role in many processing phases of fingerprint recognition such as enhancement, minutiae extraction and matching. Therefore, accurate estimation of orientation filed is much necessary. In this paper, a novel PDE-based method is proposed for regularization of orientation field for low-quality fingerprint images. The method consists of four steps. Firstly, the coarse orientation field is computed using traditional gradient-based approach. Secondly, the reliability map of the orientation field is computed based on a procedure of multiscale coherence analysis. Then the orientation in the low-reliable region is reconstructed with the surrounding data by means of image inpainting technique. Finally, nonlinear diffusion filtering with adaptive diffusivity is performed on the whole orientation field. Experiments on the NIST SD4 fingerprint database indicated that the proposed algorithm is capable to estimate the orientation field accurately, especially for poor-quality fingerprints, and it can be integrated into fingerprint recognition systems to improve the performance.