Understanding human bodies and their psychological, neural & computational contributions to identifying people
Dr. Alice O'Toole, University of Texas, Dallas
Monday, October 5th at 5:30 p.m., 118 Psychology
Abstract
Person recognition in natural viewing conditions often begins at a distance. Because the identity-specific information in bodies is visually coarse, identity perception from the body can be robust and accurate even in suboptimal viewing conditions (e.g., poor illumination, low resolution). Although much is known about face recognition, remarkably little is known about the role of the body in person identification. This lack of understanding can be traced to limited data aimed at some fundamental and relevant questions about bodies and body perception. How does body shape vary across individuals? How do we describe bodies? How do we combine information from the face and body for recognition? In the first part of the talk, I will cover experiments comparing state-of-the-art, computer-based face recognition systems to human performance. These experiments have pointed to people’s use of bodies for identification when the viewing conditions make recognition from the face challenging or impossible. The experiments also show some surprising ways in which body processing for the purpose of recognition escapes conscious awareness. In the second part of the talk, I will begin with a simple description-based approach to quantifying the information in human body shapes. In this work, people rated the applicability of body shape descriptor terms (e.g., stocky, curvy, built) for a large number of people. Correspondence analysis (Benzécri, 1973) was applied to these data to create a similarity space that contained both the bodies and the descriptor terms. We mapped between the perceptual space and a “physical” similarity space that was created from laser scans of bodies (cf., Anguelov et al. 2005). We found remarkable parity between the two spaces. Reconstructions of the original bodies in the physical space resembled the bodies on which they were modeled. The results indicate that language based descriptions of bodies can be used to reverse engineer a space that captures physical variations in real bodies. In the third part of the talk, I will discuss an fMRI experiment that illustrates the contribution of bodies to person recognition.
Benzécri, J.-P. (1973). L'Analyse des Données. Volume II. L'Analyse des Correspondances. Paris, France: Dunod
Anguelov, D., Srinivasan, P., Koller, D., Thrun, S., Rodgers, & Davis, (2005). SCAPE: Shape Completion and Animation of PEople. ACM Trans. Graph. (Proc. SIG- GRAPH) 24, 3, 408–416.
Suggested Reading
Allyson Rice, P. Jonathon Phillips, Vaidehi Natu, Xiaobo An, and Alice J. O'Toole (2013). Unaware person recognition from the body when face identification fails. Psychological Science.[.pdf]
David White, P. Jonathon Phillips, Carina A. Hahn, Matthew Hill, and Alice J. O'Toole (2015). Perceptual expertise in forensic facial image comparison. Proceedings from the Royal Society, 282.[.pdf]
Vaidehi Natu, David Raboy, and Alice J. O'Toole (2010). Neural correlates of own- and other-race face perception: Spatial and temporal response differences. Neuroimage, 4C.[.pdf]