In the context of the appearance-based paradigm for object
recognition, it is generally believed that algorithms based on LDA (Linear
Discriminant Analysis) are superior to those based on PCA (Principal Components
Analysis). In this communication, we show that this is not always the case