Abstract. In this paper we consider to study the distribution of the vectors of face in the dimensional space (n x m pixels of the image), and we have developed a face recognition that works under varying pose dealing with N different individual given under M different view/ poses and illumination. We construct an automatic algorithm that computes and finds clusters within the training group preserving intrinsic human face characteristics. The algorithm named K-PCA applies a SOM neural network in the cluster stage and applies the PCA method into each cluster, that is, each cluster forms an eigenface.
[ project home page | contact authors ]