Abstract
In an earlier work we hypothesized that face recognition processing may undergo a qualitative shift at around a critical size of 4-5 degrees, with expert face processing dominating recognition at larger sizes. This was based on our results showing that critical spatial frequencies used for face recognition increase with size for faces smaller than 4.7 degrees but stabilize at a lower relative frequency for larger faces (Oruc & Barton, 2011). In this study we explicitly tested this hypothesis by measuring contrast thresholds for recognizing upright and inverted faces at sizes between one and ten degrees of visual angles. We computed recognition efficiencies by comparing human data to the performance of an ideal observer on the same task. Our modified "CSF-ideal" observer model is limited by the same visual acuity and sensitivity constraints as our human observers, rendering the performance of the CSF-ideal observer dependent on stimulus size (Oruc & Landy, 2009). Thus our CSF-ideal incorporates any decreases in recognition performance that are solely due to visibility constraints of the human observers introduced by size. The pattern of upright face recognition efficiencies showed a step-function-like profile with efficiencies doubling abruptly at 5 degrees per face width. On the other hand, efficiencies for inverted faces remained largely flat across the size range tested. The critical size at which the shift for upright faces occurs corresponds to viewing a face at around one and a half meters in the naturalistic setting. These results show that face size is a critical factor in recognition performance. Although not conclusive, they also suggest that in upright-face perception distinct recognition processes are involved depending on size, with expert processes dominating sizes typical of close-range social interaction.
Meeting abstract presented at VSS 2012