Matthew Pachai


VSS Best Poster Awards 2013

Masking of individual facial features reveals the use of horizontal structure in the eyes.

Matthew V. Pachai1, Allison B. Sekuler1,2, Patrick J. Bennett1,2; 1-Department of Psychology, Neuroscience & Behaviour, McMaster University, 2-Centre for Vision Research, York University.

We previously demonstrated that information in the horizontal band is maximally diagnostic for face identification, and the extent to which observers preferentially utilize this information is correlated with their face identification accuracy (Pachai et al., VSS 2011). However, it remains unclear how this diagnostic information is distributed across the face, and from which regions observers extract this information. The present experiment addressed these questions using a 10-AFC face identification task in which stimuli were masked with localized patches of white, horizontal, or vertical noise at one of four rms-contrast levels (0.01, 0.1, 0.2, and 0.3) centred on one or more face parts (left eye, right eye, nose, and mouth).

The various noise types and contrasts, plus a no-mask condition, were intermixed randomly within sessions, and the various mask locations were blocked across sessions. A template-matching simulated observer demonstrated more masking with noise centred on either eye than the nose or mouth, and more masking with horizontal than vertical noise centred on either eye or the mouth, supporting the idea that the eyes are maximally informative for face identification, and that the eyes and mouth contain diagnostic horizontal structure. In human observers, masking was negligible when noise was centred on any single feature regardless of noise contrast. Observers also demonstrated a minimal threshold increase when we simultaneously masked the nose/mouth or one eye/nose/mouth. However, thresholds were elevated signifi cantly at multiple noise contrasts when both eyes were masked simultaneously, and greater masking was obtained with horizontal than vertical noise.

Efficiency relative to the template-matching observer also was higher when at least one eye was unmasked than when both eyes were masked. Together, these results suggest that human observers preferentially utilize diagnostic horizontal structure contained in the eye/eyebrow region to identify upright faces, and that performance suffers when this information is rendered unavailable.

Acknowledgement: NSERC, Canada Research Chair Programme

Click on image to view a PDF of the winning poster.