Most of the work on face processing has relied on either feature-based or template-based techniques . Pentland et al.  has addressed the face processing problem by using eigentemplates. A view-based multiple-observer eigenspace technique is used for face recognition under variable pose. In addition, a modular eigenspace description technique is used which incorporates salient features such as the eyes, nose and mouth, in an eigenfeature layer. This method also provides an automatic feature extraction mechanism.