CDVSec: Privacy-Preserving Biometrical User Authentication in the Cloud with CDVS Descriptors

We explore the use of CDVS-compressed SIFT descriptors for user authentication via fingerprint matching over the cloud. Previous research has shown that SIFT descriptors are suitable for fingerprint matching, however they do not guarantee user-privacy nor they enable efficient storage and matching. The MPEG CDVS standard is devised to extract compressed SIFT descriptors from natural images and compare them in the compressed domain. CDVS descriptors can potentially prevent recovering the original fingerprints from the compressed descriptor, strengthening user privacy. Also, CDVS descriptors do enable efficient storage and matching, a key factor with bandwidth-constrained mobile devices. We acquire a database of raw fingerprint images and we experiment with different acquisition, compression and matching schemes. Experiments show that CDVS descriptors yield a large number of repeatable, highly distinctive, keypoints from fingerprint images. Proper control over the encoding and matching algorithms shows low false acceptance rates and high true acceptance rates. Finally, we demonstrate the use of CDVS descriptors for cloud-based user authentication experimenting with a mobile client and a remote verification server over a realistic testbed.