Automatic face classifications by self-organization for face recognition

We propose a method of face recognition that can consistently identify every face angle, assuming it is used in open spaces such as a normal room. We obtain the learning images not from an ideal world but from the real world, where users can move around freely with no constraints. We then automatically classify the face images that vary according to the user´s position and posture by self-organization (unsupervised learning), and create a discrimination circuit using only the best face images for the recognition task. We show that the recognition rate for images with various facial angles in the real world can be improved by automatic classification through self-organization.