We propose a face recognition method that can identify stably for every face angle, assuming usage in large spaces such as a normal room. In our method, we obtain the learning images not from the ideal world but from the real world, in which users freely move around with no constraint. Some examples are shown below.

We automatically classify the face images that vary according to the user's positions and postures by self-organization (unsupervised learning), and created the discrimination circuit using only advantageous face images for recognition.

A example of self-organization map is shown in right. We showed that the recognition rate for various face angle images in the real world was improved by automatic classification by Self-Organization.