Invariant representation of orientation fields for fingerprint indexing

Orientation fields can be used to describe interleaved ridge and valley patterns of fingerprint image, providing features useful for fingerprint recognition. However, for tasks such as fingerprint indexing, additional image alignment is often required to avoid confounding effects caused by pose differences. In this paper, we propose to employ a set of polar complex moments (PCMs) for extraction of rotation invariant fingerprint representation. PCMs are capable of describing fingerprint ridge flow structures, including singular regions, and are tolerant to spurious orientations in noisy fingerprints. From the orientation fields, a set of rotation moment invariants are derived to form a feature vector for comprehensive fingerprint structural description. This feature vector gives a compact and rotation invariant representation that is important for pose-robust fingerprint indexing. A clustering-based fingerprint indexing scheme is employed to facilitate efficient and effective retrieval of the most likely candidates from a fingerprint database. Our experimental results on NIST and FVC fingerprint databases indicate that the proposed invariant representation improves the performance of fingerprint indexing as compared to state-of-the-art methods. ► A set of PCMs is investigated to reconstruct fingerprint orientation fields. ► Exploit PCMs to generate rotation moment invariants as the indexing feature. ► Experiments show effectiveness of the invariant features for fingerprint indexing.