Purchase this article with an account.
Ken Nakayama; Face specific processing: role of local features in an affine metric. Journal of Vision 2003;3(9):91. doi: 10.1167/3.9.91.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To understand the nature of face specific holistic/configural processing, we compared the effect of narrow band masking noise on the discrimination of upright and inverted faces. Signal to noise thresholds in a four-alternative forced-choice task with faces of varying lighting and constant pose showed that face processing is most degraded by noise of approximately 12 cycles/face, with a much broader masking function for inverted as opposed to upright faces. As such, face-specific processing (the ratio of upright to inverted masking) preferentially samples in this spatial frequency band. Using notch-filtered noise, having all spatial frequencies except those most important for face specific encoding, we created a stimulus which selectively isolates face processing and shows a consistent two-fold difference in threshold for upright vs inverted faces. In the presence of notch filtered noise, face recognition thresholds are largely unaffected by affine image distortions but are significantly elevated by a non-linear spatial distortion. Conclusion: The peak spatial frequency of the face specific processing at 12 cycles/face indicates the importance of local facial features and the effects of spatial distortion are consistent with an affine metric.
This PDF is available to Subscribers Only