The human face is one of the most important visual stimuli we encounter in our daily lives. A face can tell us who someone is, how that person feels, and what intent they may have at the moment we see them. Face perception has been a focus of our research for many years. In particular, we have been interested in how the human visual system represents the complex identity and social information in faces.
We are collaborating with researchers at Johns Hopkins University on the analysis of the visual representations of faces that emerge state-of-the-art face recognition algorithms. These algorithms operate effectively across images that vary in illumination, viewpoint, and image quality), and now perform at levels that surpass untrained humans and compete with skilled, professional face examiners.
Our research compares the accuracy of professional forensic examiners, untrained people, and state of the art face recognition systems based on deep convolutional neural networks. This work build on previous humans-machine comparisons carried out in collaboration with researchers at National Institute of Standards and Technology (NIST). Over the past decade, we have published human benchmarks for state-of-the-art machine-based face recognition systems.
The human body can be described in many ways, but simple combinations of words can elicit vivid mental images of complete body shapes. The “stout, portly gentleman”, “the lean, lanky athlete”, and the “shapely hour-glass lady” are all easy to imagine. We have explored the relationship between word based descriptions and full scale 3D body models, in collaboration with researchers at the Max Planck Institute for Intelligent Systems.
How do we recognize people we know? Successful recognition can be achieved using identity information from faces, bodies, and from the natural “biological motions” of a person. We study how people make use of this diverse information to recognize someone. Behaviorally, we examine the spatiotemporal course of recognition over variable distances.
In this project, we are testing the ability of professional forensic facial examiners in face identification tasks and comparing their accuracy to untrained human participants and computer algorithms.
Functional neuroimaging techniques, including fMRI, have made it possible to “observe” the activity of the human brain as a person engages in a task. Our work focuses on expanding the types of tools used to analyze fMRI data on face perception, using pattern classification algorithms. We have used these tools to study how the brain codes faces, bodies, and people in motion and at rest. We have also looked at the coding of person familiarity in naturalistic viewing conditions.