Create extra-small biometrical footprints for face recognition, capable of being contained in low-density QR codes with high recognition accuracy. This would allow for validating connectionless physical assets with direct face image recognition without storing or connecting to external services.
Design of a new architecture of clustered AI-based autoencoders that are trained with large face images datasets. New loss functions are designed using vectorized dimensions.
(1) Accurate dimensionality reduction of face features up to four dimensions for enhanced low-density biometric QR codes; (2) a topology-aware AI-based novel architecture based on multiplexing autoencoders over clusterized data, applied to biometrical face recognition; (3) a case study based on biometrical QR recognition using the SAAV architecture; and (4) an experimental implementation of the case study using a large face dataset.
The results are solid, providing a compression rate of 80 times (1/80) over Google’s FaceNet architecture, with 89% of recognition success. It should recall attention not only from the academics but also from the real business as it is a solid improvement with practical application.
The technology can be directly transferred to multiple business applications: – transport tickets (plane, train, etc.) – events tickets (concerts, festivals, etc.) – self-contained biometrical stickers for physical assets such as computers, cars, etc. – self-contained personal identifications for emergencies – personal transport cards for logistics and many more.
It could have a great implication when applied to emergency situations in refugee camps, or hospitals, to create self-identifying bracelets or cross-identifying bracelets for mom’s and babies.
The techniques exposed are 100% original.
