Demonstrator for Privacy-Preserving Face Recognition

Bachelor Thesis

finished


Description

In order to improve public safety, government and private security companies increasingly put on video surveillance, which threatens the privacy of the public. In this Bachelor-Thesis we introduce a new demonstrator for privacy-preserving face recognition. The demonstrator performs privacy-preserving face recognition on a client input image with a database of images which outputs whether there is a match but reveals no other information to any party. This scenario has multiple applications, such as a camera-based surveillance of public places or identification at border control. Nowadays a camera-based surveillance of public places is a security feature, which can lead to a higher security but also reduces the privacy of the public. Therefore, we propose a system that uses secure two-party computation which additionally protects privacy of the public. Simultaneously, the face recognition needs to be faster to make them applicable to real-world scenarios.For this task SCiFI (SSP’10), a system for Secure Computation of Face Identification can be used. But the recognition part of SCiFI is originally done by computing the hamming distance based on homomorphic encryption on vectors, each representing an image, what has a high runtime and communication complexity. We replaced this part by a secure hamming distance computation using the ABY-Framework (NDSS’15), which results in a performance improvement. ABY is a mixed-protocol framework for secure two-party communication, which allows to combine secure computation schemes and efficiently convert between them. The hamming distance computation is based on boolean circuit and can be evaluated with Yao’s garbled circuit-Protocol or with the GMW-Protocol. The feature extraction of faces is still based on SCiFI, since it is well designed for secure computation and has a high recognition performance. The evaluation of the circuit for the hamming distance is very fast, so the new Demonstrator is a significant improvement in runtime compared to the original SCiFI system.

Publication: Nils Schroth - Demonstrator für privatsphäre-schützende Gesichtserkennung (German)


Start: 18.04.2016

Supervisor:

Research Areas: ENCRYPTO



A A A | Drucken Print | Impressum Impressum | Sitemap Sitemap | Suche Search | Kontakt Contact | Webseitenanalyse: Mehr Informationen
zum Seitenanfangzum Seitenanfang