FACE RECOGNITION - THE PROBLEM OF COMPENSATING FOR CHANGES IN ILLUMINATION DIRECTION

A face recognition system must recognize a face from a novel image des pite the Variations between images of the same face. A common approach to overcoming image Variations because of changes in the illumination conditions is to use image representations that are relatively insens itive to these variations. Examples of such representations are edge m aps, image intensity derivatives, and images convolved with 2D Gabor-l ike filters. Here we present an empirical study that evaluates the sen sitivity of these representations to changes in illumination, as well as viewpoint and facial expression. Our findings indicated that none o f the representations considered is sufficient by itself to overcome i mage variations because of a change in the direction of illumination. similar results were obtained far changes due to viewpoint and express ion. Image representations that emphasized the horizontal features wer e found to be less sensitive to changes in the direction of illuminati on. However, systems based only on such representations failed to reco gnize up to 20 percent of the faces in our database. Humans performed considerably better under the same conditions. We discuss possible rea sons for this superiority and alternative methods for overcoming illum ination effects in recognition.