9 Conclusion
Privacy-preserving computation offers the promise of obtaining results dependent on private data without expos-ing that private data. The main drawback is that current pro-tocols for privacy-preserving computations are very expen-sive and impractical for real-scale problems. In this work, we have shown that those costs can be substantially reduced for a large class of biometric matching applications by de-veloping efficient protocols for Euclidean distance, finding the closest match, and retrieving the associated record. Our approach involves using the normal by-products of a gar-bled circuit evaluation to enable very efficient oblivious in-formation retrieval, and we believe this technique can be ex-tended to many other applications. Our experimental results support the hope that privacy-preserving biometrics are now within reach for practical applications.