Insights into human face recognition processing strategies through feature-based quantitative comparisons

Matt Peterson

Psychology Department

UC Santa Barbara

Date: Friday, May 23, 2008
Place: Harold Frank Hall, 4164
Time: 2:00 pm — 3:00 pm


Abstract:

Humans are said to be experts at recognizing faces. Within a few hundred milliseconds we are able to automatically discern a person’s identity through a quick glance of the face across an immense range of lighting conditions, pose locations and orientations, occlusions and contexts. Considering the difficulty that even modern face recognition algorithms encounter, the human brain’s strategy seems well-optimized for this important task. Here, we use statistical modeling to estimate the amount of discriminating visual evidence within specific features of a large database of frontal view face photographs. We compare these objective results to human identification performance when various features are removed. The efficiency with which human observers utilize the visual information in certain features gives us a window into the brain’s image processing strategy. Aside from being an interesting scientific query, these results may aide future automated face recognition algorithms.

MATT PETERSON is a member of the Vision and Image Understanding Laboratory and a PhD candidate in the Psychology Department at UCSB. His research advisor is Prof. Miguel Eckstein. He was funded by the digital multimedia IGERT grant from Fall 2005 through Summer 2007.