A morphable model for the synthesis of 3D faces
Proceedings of the 26th annual conference on Computer graphics and interactive techniques
Pattern Classification (2nd Edition)
Pattern Classification (2nd Edition)
Journal of Cognitive Neuroscience
ACM Transactions on Applied Perception (TAP)
| Hi-index | 0.01 |
This paper presents a psychophysical analysis of the discrimination thresholds of human faces that are varied along different directions in Face Space. Generated by a 3D Morphable Model, the stimuli are frontal views of original laser scans that are modified in shape, texture or both. Discrimination thresholds are then measured in a four Alternative Forced Choice (4AFC) design and a staircase method. In Experiment 1, texture and shape are varied separately along a set of principal component directions. For different components, the results show a consistent pattern of high and low thresholds across individual participants. We compare Mahalanobis distance, Euclidean distance in face space, and 2D image differences as potential predictors for human discrimination thresholds. The goal of Experiment 2 is to investigate the interaction between shape and texture. The stimuli include combined and separate variations of shape and texture in the 4AFC staircase setup, which are analyzed in a within-subject paradigm. The results indicate that participants rely on both shape and texture for their decision. The experiments help to assess different potential models of the fusion of shape and texture cues, and provide useful information for computer graphics applications such as character design interfaces.