Neuromorphic vision networks for face recognition

Abstract

Abstract This chapter proposes a new hardware-based implementation of real-time face-recognition application that is inspired by the cortical neuron firing in the human brain. The face-recognition method presented here can detect and store the most relevant data at the sensor. Then, in contrast to the software approaches, it uses the proposed memristive threshold logic network of reconfigurable cells to build a fast, robust, parallel and scalable architecture for storage and detection of most relevant pixel information. The main idea is that the memristive threshold logic cells proposed here can detect the most relevant and significant pixel intensity variations at the level of sensors that receive video or image input. Then the storage of these data occurs depending upon the memristive state of a memristor. The proposed neuromorphic face-recognition system shows the benefits of on-chip density, low power dissipation and scalability when used for high resolution networks.