A principal component neural network-based face recognition system and ASIC implementation

Principal component analysis (PCA) finds wide usage in computer-aided vision applications and one such application is face recognition. The neural network that performs PCA is called a principal component neural network (PCNN). This paper presents a new PCNN-based face recognition system. The proposed recognition system can tolerate local variations in the face such as expression changes and directional lighting. An optimal digital hardware design is proposed for PCNN. An ASIC implementation of the proposed design yields a throughput of processing about 11,000 inputs per second during the training phase and about 19,000 inputs per second during the retrieval phase. The customized hardware-based recognition is about 10⁵ times faster than a software-based recognition in a PC. Such results are valuable for high-speed applications.