Towards a Face Verification System-on-Card with Compound-Eye Camera

Electronic passports and other identity documents store biometric reference data of the document holder (such as a face image and fingerprint images) in a smart-card chip with contactless interface to enable the automatic verification that the holder is the legitimate holder. Off-card biometric comparison involves sending sensitive biometric reference data via near-field communication to a background system. Even if encrypted, the transfer of biometric data requires considerable trust in the background system and can threaten the privacy of the cardholder. A biometric system-on-card integrating the biometric sensor and signal processing into the card enhances privacy because sensitive biometric data never leave the card. System-on-card technology with capacitive fingerprint
sensors integrated into a smart card is already mature. Optical camera modules for capturing face images, however, have not been integrated into smart cards yet. Ultra-thin camera modules inspired by the compound eyes of insects could be integrated into smart cards, but would the computing resources on a smart card be sufficient for face recognition? This paper assesses the feasibility of a face-verification system-on-card using an ultra-thin compound-eye camera and a pre-trained lightweight neural network. Taking the resource limitations of smart cards into account, a face verification pipeline from capture to verification decision is outlined, and its accuracy is estimated. This paper shows that lightweight face verification using a compound-eye camera is feasible with acceptable
accuracy but requires more memory and computing power than available on today’s off-the-shelf smart cards.