Skip to content

Back to the catalog

GhostFaceMPC: Privacy-Preserving Face Recognition with Multi-Party Computation
eprint.iacr.orgpaper

GhostFaceMPC: Privacy-Preserving Face Recognition with Multi-Party Computation

GhostFaceMPC is presented as the first end-to-end face recognition system that runs both the deep CNN feature extractor and the matching step under secure multi-party computation. Raw images and biometric embeddings are never exposed at any stage, from feature extraction through storage to matching. The authors contribute an MPC-friendly quantization of the GhostFaceNet model and circuit optimizations that make this practical, and they release code on GitHub. It is relevant as an example of making CNN inference feasible under cryptographic constraints.

TL;DR First end-to-end MPC face recognition: MPC-friendly quantized GhostFaceNet plus circuit optimizations for private CNN inference and matching.
Open original ↗
Also seen from another angle🛡️ Security →
Document

What it claims

GhostFaceMPC runs the whole face recognition pipeline under secure multi-party computation: a deep CNN feature extractor (GhostFaceNet), storage of templates, and matching. Neither raw images nor biometric embeddings are revealed at any stage. The authors call it the first end-to-end system to put both the CNN extractor and the matching step under MPC. The page gives no accuracy, latency, communication or party-count figures.

How it works

The abstract names two contributions and does not describe either:

  • An MPC-friendly quantization of GhostFaceNet. The bit-width, the quantization scheme and the accuracy cost are not stated.
  • Circuit optimizations that the authors say make face recognition under MPC "practical." Which circuits, which MPC protocol, and the security model (number of parties, semi-honest or malicious) are not stated.

How to use it

Code is at https://github.com/crypto-unict/GhostFaceMPC. The page gives no install or run instructions, and it does not state the code's licence. Check the repository directly before adopting it.

The authors are Davide Carnemolla, Dario Catalano and Mario Di Raimondo, from the University of Catania (unict.it). The ePrint category is Applications.

Limits

  • Abstract only. The fetched page has the abstract and metadata, nothing more. Nothing here supports "practical" with numbers, so read the PDF for runtime, bandwidth, accuracy against plaintext GhostFaceNet, and the threat model before relying on the claim.
  • Not peer-reviewed. The paper is listed as a preprint, received and approved on 2026-10-15.
  • Paper licence. The paper is CC BY-NC-ND. That allows no commercial use and no derivatives of the text. It covers the paper, not necessarily the code, whose licence is unstated on this page.
  • Scope of "practical" is unknown. No deployment setting, gallery size or hardware is given. It is unclear whether this suits 1:1 verification only or also 1:N identification at scale.

Organisations named here

  • University of Catania

Technologies named here

  • GhostFaceMPC
  • GhostFaceNet
  • CNN
  • GitHub
  • secure multi-party computation

Techniques named here

  • MPC-friendly quantization of GhostFaceNet
  • Circuit optimizations for MPC

Related in research