A Self-tuning People Identification System from Split Face Components

Multimodal systems can solve a number of problems found in unimodal approaches. We experimented going further along this line, by dividing the face into distinct regions (components) and processing each of them within a single subsystem. Such subsystems are then embedded in a more complex multicomponent architecture. In this way, typical tools of multimodal systems, such as reliability margins or fusion schemes, can be usefully extended to the single face biometry. An additional innovation element in this work is the definition of a global system auto-verification and auto-tuning policy able to produce a significant accuracy enhancement. The paper explores three integration architectures with different subsystem interconnection degree, demonstrating that a tight component interaction increases system accuracy and allows identifying unstable subsystems.