Image primitives supporting perception of animate forms

Abstract

The human visual system can recognise familiar forms, most notably faces, in other objects or patterns, a phenomenon known as pareidolia. The patterns that elicit pareidolia range from meaningful to ambiguous and random images, making it hard to generalise across the featural or configurational properties that trigger different types of pareidolia. Here, we aim to characterise the minimal stimuli associated with different types of pareidolia and investigate the extent to which pareidolia is tuned to variations in natural scene statistics and symmetry. Participants in the current study viewed a range of synthetic noise patterns varying in their spatiotemporal spectral and symmetry characteristics and reported any shapes or structure perceived in these patterns. The patterns with spatiotemporal properties typical of natural scenes generated the highest number of responses with more animate, rather than inanimate, forms overall. While faces were the most reported animacy-related percept, responses covered a wide range of animate agents including animals and mythical creatures. The greatest number and the highest proportion of animacy-related percepts were observed in vertically symmetrical patterns compared to other types of pattern symmetry. Together, the current study establishes that pareidolia is tuned to natural scene statistics and biased towards animate forms, especially in patterns with vertical symmetry. It also demonstrates the usefulness of synthetic noise stimuli for pareidolia research.