Abstract
In this study, the challenging task of unfamiliar face recognition is investigated, focusing on the effects of viewpoint and orientation (together referred to hereafter as pose, and individually as yaw and roll). Two experiments were conducted using singly, briefly presented stimuli in a sequential matching task. In experiment 1, the face pose presented in the study display was randomized from 84 (the combinatorial product of 7 yaw angles and 12 roll angles). Test pose was always identical to that used in the study display. In signal trials, the face shown in the test display matched that shown in the study display; in noise trials, the face shown in the test display was new. In experiment 2, study and test pose were individually randomized (i.e. typically did not match), requiring observers to recognize faces presented at different poses from study to test. Experiments 1 and 2 were otherwise identical. Four observers completed 16800 trials per experiment. This experimental framework yields performance data without prejudice for specific viewpoint effects, enabling a range of hypotheses to be tested post hoc. In this study, the ¾; view advantage, qualitative aspects of the face inversion effect, and the impact of face size, shape, and the impact of low-level image statistics on face recognition are evaluated and correlated with face pose. Results may be summarized as follows. Continuous pose-contingent performance was observed across both rotational dimensions. Pose at both study and test was seen to be significant (experiment 2). Both absolute pose and pose offset were seen to be significant (experiment 2). Limited evidence for a ¾; view advantage in experiment 2 only was found. Image statistics were seen to be exploited to a greater degree during the recognition of inverted faces, but contributed significantly to recognition in all pose conditions (experiments 1 and 2).