Online Incremental Face Recognition System Using Eigenface Feature and Neural Classifier
Seiichi Ozawa, Shigeo Abe, Shaoning Pang, Nikola Kirilov Kasabov · 2009
This chapter described a new approach to constructing adaptive face recognition systems in which a low-dimensional feature space and a classifier are simultaneously learned in an online way. To learn a useful feature space incrementally, we adopted Chunk Incremental Principal Component Analysis in which a chunk of given training samples are learned at a time to update an eigenspace model. On the other hand, Resource Allocating Network with Long-Term Memory (RAN-LTM) is adopted as a classifier model not only because incremental learning of incoming samples is stably carried out, but also because the network can be easily reconstructed to adapt to dynamically changed eigenspace models. To evaluate the incremental learning performance of the face recognition system, a selfcompiled face image database was used. In the experiments, we verify that the incremental learning of the feature extraction part and classifier works well without serious forgetting, and that the test performance is improved as the incremental learning stages proceed. Furthermore, we also show that Chunk IPCA is very efficient compared with IPCA in term of learning time; in fact, the learning speed of Chunk IPCA was at least 8 times faster than IPCA.