Practical identification with encrypted biometric data using oblivious RAM

Bringer et al. introduced a new primitive that enables identification over encrypted biometric data stored on a remote database and that preserves the privacy of the database users toward the database holder. The tools used in their scheme however rely on heavy cryptographic protocols that prevent it from being used in practice for large databases. We modify their technique using Oblivious RAM and prove its usability by an implementation adapted to iris recognition. For a database of 250,000 IrisCodes, only 3.6 MB of data need to be locally stored and an identification request using our new protocol requires less than 350 ms of computation time.