A Real-Time Personal Authentication System with Selective Attention and Incremental Learning Mechanism in Feature Extraction and Classifier

摘要

We propose a new approach for a real-time personal authentication system, which consists of a selective face attention model, incremental feature extraction, and an incremental neural classifier model with long-term memory. In this paper, a face-color preferable selective attention combined with the Adaboost algorithm is used to detect human faces, and incremental principal component analysis (IPCA) and resource allocating network with long-term memory (RAN-LTM) are effectively combined to implement real-time personal authentication systems. The biologically motivated face-color preferable selective attention model localizes face candidate regions in a natural scene, and then the Adaboost based face detection process identifies human faces from the localized face-candidate regions. IPCA updates an eigen- space incrementally by rotating eigen-axes and adaptively increasing the eigen-space dimensions. The features extracted by projecting inputs to the eigen-space are given to RAN-LTM which learns facial features incrementally without unexpected forgetting and recognizes faces in real time. The experimental results show that the proposed model successfully recognizes 200 human faces through incremental learning without serious forgetting.