Abstract
A match-to-sample paradigm was employed to investigate various factors that can influence the discrimination of facial identity. Each trial began with a brief presentation of a “sample” face with a neutral expression. This was followed in sequence by a pattern mask, and a pair of new images: A target image of the same individual as depicted in the standard but with a different facial expression and/or direction of illumination; and a foil image of a different individual. Observers were required to identify the target face by pressing an appropriate response key as quickly as possible. The relative similarity of the standard to the target and foil in terms of pixel intensities or Gabor filter outputs was systematically manipulated across trials. The results revealed that the relative similarity of the standard to the target and foil had a negligible effect on the speed and accuracy of the observers' responses, although there were significant differences in performance for different facial expressions.
In an effort to compare observers' performance with a typical image-based model of face recognition, a PCA representation was computed from a large set of facial images different from those used in the behavioral study. This representation was then used to perform the match-to-sample task described above on the same stimuli images used with subjects. Although the overall accuracy of the model was similar to that of human observers, the correlation of the human and model performance across the different experimental conditions accounted for less than 5% of the variance.