Perceived similarity in face measurement

• Automatic face recognition involves decisions based on face measurement results.
• Possible improvements may arise from investigating the reliable human face recognition process.
• An investigation of perception of similarity among faces, focused on two recognition procedures is presented.
• Original methods are applied to direct measurement of perceptual quantities.

Automatic face recognition is a widely used biometric identification method, since it is based on standard video surveillance systems and does not require the collaboration of the person to be recognised. The recognition process involves decisions based on the face measurement results. The reliability and robustness of the recognition procedure are critical factors, strongly affected by influence quantities, such as face-camera relative orientation and position, illumination and expression. To improve performance it makes sense to imitate the recognition process in humans, who succeed in recognising faces even from poor quality images. It is therefore interesting to investigate the perception of similarity between faces, with the aim of gaining insight into the process of human face recognition. We report on an investigation that has been carried out by original methods developed in our laboratory for the direct measurement of perceptual quantities. Such methods have been implemented for two complementary recognition procedures, based either on the global appearance of a face or on the detection of specific repere points.