Benjamin Balas¹, Emily Westrick¹; ¹North Dakota State University

The detection of face-like structure in noise patterns (face pareidolia) is sensitive to the statistics of those images. In prior work, we reported that adults’ rates of reporting face pareidolia in fractal noise images depends on spectral power (steeper power spectrum coefficients reduce pareidolia rates) and orientation content (horizontal or vertical orientation filtering reduces pareidolia rates). The tuning of face pareidolia to these image statistics may reflect sensitivity to statistical regularies in natural images that develops slowly during the school-age years. For example, children only acquire an adult-like bias favoring horizontal orientation energy in face recognition tasks gradually during middle childhood. Presently, we investigated this possibility by measuring face pareidolia as a function of orientation statistics in adults and a developmental sample. We hypothesized that compared to adults, children would exhibit more consistent pareidolia rates across isotropic, horizontally-filtered, and vertically-filtered noise images. We asked children (aged 5-12) and adults to complete an unconstrained face pareidolia task in which they were shown 4 exemplars each of isotropic fractal noise textures (1/f power spectrum) and orientation-filtered versions of those images with horizontal and vertical passbands applied. Unlike our prior study examining orientation energy and pareidolia, we additionally imposed symmetry across the vertical meridian (Palas & Webster, 2013) to increase overall face pareidolia rates. Participants were given 2 minutes to view each image and circle the pareidolic faces that they saw in this interval. Contrary to our previous results, we found that while orientation filtering of both types did significantly reduce pareidolia rates in our adult sample, horizontally-filtered images yielded more pareidolic faces than vertically-filtered images. Also, consistent with our initial hypothesis, children exhibited stable rates of pareidolia across all conditions. We discuss these results in terms of the emerging template for face detection during childhood and cues that support perceptual grouping.

Acknowledgements: This work was supported by NSF grant BCS-2338600