Learning to Recognize Faces by Successive Meetings

Abstract

In this paper we focus on the face recognition problem. However, instead of following the usual approach of manually gathering and registering face images to build a training set to compute a classifier off-line, the system will start with an empty training set, i.e. no experience, and it will build it autonomously by continuous on-line learning. In that way the classifier evolves with the perceptual experience of the system, similarly to the way humans do. Experiments have been performed with 310 sequences corresponding to 80 identities. Two different configurations have been analyzed depending on the ability to detect new, i.e. unknown, identities. The results achieved evidence that if a verification stage is included the system learns fast to detect new identities. For revisitors, the accumulated error rate decreases in both cases, reaching around 50% if no verification is included. These results seem to indicate that more interaction or meetings with the different individuals are needed to affirm that their identity is familiar enough to be recognized robustly.