Chapter 8 - Face recognition by humans

Abstract

The chapter discusses the limits of human face recognition skills, cues that humans use for face-recognition, biologically plausible strategies for face recognition, and the timeline of development of human face recognition skills. Face recognition is one of the most active and exciting areas in neuroscience, psychology, and computer vision. While significant progress has been made on the issue of low-level image representation, the fundamental question of the process to encode overall facial structure remains largely open. Machine-based systems use well-designed perceptual studies that can allow precise inferences to be drawn about the encoding schemes used by the human visual system. This chapter reviews a few results from the domain of human perception that provide benchmarks and guidelines for the efforts to create robust machine-based face recognition systems. Specialized identification systems (based on novel sensors, such as close-range infrared cameras) may exceed human performance in particular settings. However, in many real-world scenarios using conventional sensors, matching human performance remains an elusive goal. Data from human experiments gives a better sense of computational strategies that could be employed to achieve it and, eventually, move past it.