Abstract
Symmetric noise patterns can give rise to strong and frequent impressions of faces along the vertical axis of symmetry (Paras et al. VSS 2006). These illusory faces invariably include a pair of eyes as the dominant feature, suggesting that a percept of eyes may be an important trigger feature for face perception. Here we examined the stimulus properties (e.g. contrast polarity or spatial scale) that can be perceptually interpreted as eyes, and how this interpretation influences the perception of other regions in the image. Stimuli were grayscale images of filtered asymmetric noise. In a subset of images a pair of Gaussian spots was added to the noise along the horizontal or vertical midline. The horizontal - but not vertical - pair causes most images to appear organized as a face. This suggests that the presence of “eyes” may engage a face processing mode leading to perceptual completion of a face, and we explore this by examining how different areas of the image are classified in the presence or absence of eyes. Interpreting the noise as a face may allow the images to be more perceptually informative and discriminable, even though the dots leading to this interpretation do not add any explicit information for the discrimination. To test this, in further experiments we compare recognition accuracy for sequences of noise images in the presence or absence of eyes.