Hashing Fingerprints for Identity De-duplication

Fraudulent identities of multiple enrollments usually link to fraud and serious breaches of law. With the vast biometric data collection, identity de-duplication has become the processing bottleneck of biometric enrollments. Recently, locality sensitive hashing (LSH) based methods have been introduced for fast retrieval of biometric identities. Most of them are working in the binary space. This paper proposes a new fingerprint indexing method based on a variant of LSH called spherical LSH (S-LSH). The proposed S-LSH based algorithm is able to hash fingerprint templates directly in the original feature space and thus avoid the intermediate step of binary transformation. In this way, S-LSH can better preserve the interpoint similarity of minutiae points. We demonstrate the effectiveness and efficiency of the new S-LSH based approach by performing fingerprint indexing experiments on the FVC2002 DBI database and comparing it with a state-of-the-art hashing based fingerprint indexing method.