Face recognition method independent of rotation and size variations

Emulating the parietal cortex, a "rotation and size spreading associative neural network" (RS-SAN net) was developed. We extended the original system to make a human face recognition system which learned and recollected human face images. Previously, the recognition characteristics of the RSSAN net were investigated for learned faces. However, the same human face images were used in both learning and recollection processes. In the present study, we investigated the recognition characteristics using different face images of the same subject. In addition, we investigated the recognition characteristics for unlearned faces. The original system was improved to make the recognition system cope with unlearned faces. We introduced an inner product and a minimum distance as new shape recognition criteria. By setting the threshold ranges of the inner product and the minimum distance as 0.999/spl sim/0.998 and 0.04/spl sim/0.06, respectively, the false rejection and the false acceptance rates became 0% in both criteria.