Benjamin Balas¹, Emily Westrick¹, Lily Roshau¹, Molly Setchfield¹; ¹North Dakota State University

Face pareidolia, the phenomenon of seeing face-like patterns in non-face images, has a dual nature: Pareidolic patterns are experienced as face-like, even while observers can recognize the true nature of the stimulus (Stuart et al., 2024). Though pareidolic faces seem to result largely from the canonical arrangement of eye spots and a mouth, we hypothesized that competition between veridical and face-like interpretations of pareidolic patterns may constrain face pareidolia in natural scenes and textures. Specifically, we predicted that contrast negation, which disrupts multiple aspects of mid- to high-level recognition, may increase rates of face pareidolia in complex natural textures by weakening the veridical, non-face stimulus interpretation. We presented adult participants (N=27) with a series of natural images depicting textures like grass, leaves, shells, and rocks. Each of our original 8 grayscale images was first used to create a mirror-symmetric stimulus (per Paras & Webster, 2013) by dividing pictures at the vertical midline and replacing the right half of the image with a flipped copy of the left half. These symmetric images were then rendered in positive and negative contrast for a total of 16 images, which were shown in a pseudorandomized order to each participant on an iPad. We asked participants to circle any patterns in each image that looked face-like, with no constraints on response time or pattern size, position, and orientation. We analyzed the resulting count data (number of circled patterns per image) with a generalized linear model including image contrast (positive or negative) as a fixed effect and participant ID and item number as random effects. This revealed a significant effect of image contrast, with contrast-negated images yielding more pareidolic face detections than positive images. We conclude that disrupting veridical object and texture recognition enhances pareidolia by compromising half of the dual nature of a pareidolic pattern.

Acknowledgements: This research supported by NSF-2338600 awarded to BB.