Face recognition independent of facial expression through SOM-based classifiers

In this paper, we evaluate four pattern classifiers built from the self-organizing map (SOM), a well-known neural clustering algorithm, in the recognition of faces independent of facial expression. The design of two of the classifiers involves post-training procedures for labelling the neurons, i.e. no class information is used prior to the training phase. The other two classifiers incorporate class information prior to the training phase. All the classifiers are evaluated using the well-known Yale face database and their performances compare favorably with standard neural supervised classifiers.