Abstract
We present an extension of the multidimensional face space based on face silhouettes (Davidenko, 2007, 2009) to include front-view faces. The front-view face space was constructed by manually placing 85 identifiable keypoints on the outline and features of 190 front-view faces compiled from 4 different databases. The coordinates of the keypoints were normalized and entered into a principal components analysis to produce orthogonal dimensions that capture the variability of front-view faces. We confirmed the validity of these face stimuli with a celebrity recognition task. In two drawing studies, we tested naive observers' ability to draw these parametric faces. In Study 1, 15 participants copied 8 upright and 8 inverted faces (in semi-randomized order) using a stylus on a touch screen, with 90 seconds to copy each face. The accuracy of each drawing was assessed by manually placing 85 keypoints on each drawing and measuring pairwise distances from the corresponding 85 keypoints on the original stimuli. We found that, contrary to common conception, participants were significantly more accurate when copying upright faces than inverted faces, suggesting holistic processing may actually aid, rather than hinder, face drawing. In Study 2, we compared 11 participants' ability to copy a face and then draw the same face from memory. Surprisingly, error rates did not differ significantly between the copying and memory conditions, demonstrating that observers are able to accurately draw parameterized faces from memory. We discuss implications of these findings for improving methods of eyewitness face reconstruction.
Meeting abstract presented at VSS 2016