Abstract
We present an efficient matching protocol that can be used
in many privacy-preserving biometric identification systems
in the semi-honest setting. Our most general technical contribution
is a new backtracking protocol that uses the byproduct
of evaluating a garbled circuit to enable efficient
oblivious information retrieval. We also present a more efficient
protocol for computing the Euclidean distances of
vectors, and optimized circuits for finding the closest match
between a point held by one party and a set of points held
by another. We evaluate our protocols by implementing a
practical privacy-preserving fingerprint matching system.