Spike-based learning rules for face recognition

This paper proposes a biologically plausible network architecture with spiking neurons for face recognition. This network consists of three parts: feature extraction, encoding and classification. Firstly, HMAX model with four layers (C1-S1-C2-S2) is used to extract face features. The proposed feature extraction method can keep selectivity invariance and scale invariance. The next important part is to encode features to suitable spike trains for spiking neural networks. In the last part, the improved Tempotron learning rule is chosen to train the spiking neural networks with reduced computation and increased fault tolerance. In order to demonstrate the performance of spiking neural networks, four databases are tested in the experiment: Yale, Extend Yale B, ORL, and FERET.