Are faces easier to recognize in 3/4 view than in full-face or profile view?
Authors: Liu, C.H., Lalonde, J. and Chaudhuri, A.
Journal: Journal of Vision
Purpose. Although the 3/4 view is often regarded as being canonical for faces, the evidence for it is rather mixed. One reason for the discrepancy may be that the results in some studies are affected by another factor-amount of angular rotation between learning and test views. We hypothesized that if this variable is controlled, the 3/4 view effect may be reduced or disappear. In addition to the general belief that 3/4 view can generalize better to different views, the 3/4 view is also believed to produce better recognition in the same view. However, the evidence for this also needs to be reconsidered, particularly if both the different-view and the same-view advantage is assessed in a single experiment. A task that involves less amount of angular rotation should be easier to perform for faces in different views. Therefore, more resources would be available to hold the same- view faces in memory. To avoid such problems, we assessed the two types of advantage in separate experiments, controlling the factors of angularrotation and view. Methods. A recognition task and a between-subject design were used. In Experiment 1, faces presented at learning and test were shown in different views. There were six conditions (full-face to 3/4, full-face to profile, 3/4 to profile, and the reverse order of these). In Experiment 2, faces presented at learning and test were always in the same view, which was either a full-face, a 3/4, or a profile view. Results. In Experiment 1, faces in 3/4 view were not recognized better either at learning or test. Only an effect of angular rotation was found in which 90 degree angular rotation produced poorer recognition performance than 45 degree rotation. In Experiment 2, all three views produced comparable results. Conclusions. Our results suggest that a better predictor of performance for face recognition in different views is the angular difference between learning and test views rather than any canonical view. For recognition in the same view, the 3/4 view is no more effective than the other views.
Preferred by: Changhong Liu