The privacy-preserving method for face recognition is homomorphic encryption, so I need the source code.

Healthcare, financial, government, and military organizations depend on encryption to secure sensitive data. Historically, this data has had to be decrypted before it could be processed or analyzed. As a result, data processing had to be performed on secured hardware, eliminating the possibility of using the cloud or other low-cost, third-party computing resources.

At the Symposium on the Theory of Computing in 2009, Craig Gentry of IBM presented a fully homomorphic encryption (FHE) scheme that made it possible to send sensitive data to an unsecured server, process it there, and receive an encrypted result, without ever decrypting the original data. While FHE was a major theoretical breakthrough, actual FHE implementations are many orders of magnitude too slow to be of practical use, particularly for large encryption keys and ciphertexts.

As a step toward a practical FHE implementation, we have developed a somewhat homomorphic encryption (SHE) scheme that, with modifications, can be converted into an FHE scheme. Current FHE implementations depend on complicated operations that are inefficient when performed on a CPU, and our goal was to take advantage of the parallelism and pipelining of FPGAs using MATLAB®, Simulink®, and HDL Coder™. Homomorphic encryption is an active area of study, and new advances are being made regularly. By using MATLAB and Simulink instead of a lower-level programming language, we can keep pace with these developments by rapidly implementing improvements to the algorithms.