Abstract
In theories of normal face recognition and in machine vision, face detection is a crucial stage. Despite its importance we know little about face detection in normal subjects or prosopagnosics. To investigate this, we compared normals and prosopagnosics on two face detection tests.
In the first test, two tone images were created by adjusting the threshold controls in Adobe Photoshop, so that darker areas became black and lighter areas became dark. Each face was surrounded by a large field of individual features drawn from other faces that served to make the facial configuration much more difficult to perceive. Blocks consisted of 48 stimuli, 36 contained a face while 12 did not. Subjects did upright and inverted blocks, and they were counterbalanced. They were asked to make a key press when they detected a face. Controls were nearly three times slower with inverted faces than with upright faces and accuracy was far worse with inverted faces.
In the second test, subject were presented with 5×5 arrays of photographs and made a key press when they detected a photograph containing a face. Thirty-seven stimuli were presented, 25 contained a face while 12 did not. It took approximately twice as long for controls to detect inverted faces compared to upright faces.
Prosopagnosics were tested with upright versions of each test. Most prosopagnosics performed normally on both face detection tests, but some showed clear impairments with both tasks. The results indicate that face detection is separate from later face processing.