Number is more than meets the eye: Unveiling segmentation mechanisms in numerosity perception with visual illusions.

Animals and humans are endowed with an adaptive ability to rapidly extract approximate numerical information from sets, yet the underlying visual mechanisms are poorly understood. Evidence suggests that visual approximate numerosity relies on segmented perceptual units, modulated by grouping cues. Indeed, perceived numerosity decreases when objects are connected by irrelevant-lines without varying low-level features. However, approximate numerosity perception has been largely studied with physical objects. Illusory contours (ICs) are crucial psychophysical tools for uncovering segmentation mechanisms built into the visual cortex. Strikingly, “illusory” objects are subjected to several perceptual biases (e.g., tilt aftereffect) akin to physical objects, indicating a common processing mechanism. Here, to unveil further similarities between real and ICs processing, we tested whether approximate numerical ICs perception is affected by connectedness grouping. In a forced-choice task, participants compared pairs of stimuli containing Ehrenstein-like ICs with varying numerosity, interspersed with four physical task-irrelevant lines. We manipulated the number of connected pairs (0, 2, or 4), aligning the lines to the gaps triggering ICs, while keeping low-level features constant across connectedness levels. Results revealed a monotonic numerosity underestimation as connections increased, and a constant precision implying a Weber-like encoding of numerosity. Furthermore, connectedness causes a proportional cost in reaction-times. These results clearly show that numerical processing of ICs ensembles is subjected to the same connectedness effect observed with real objects, suggesting a shared visual segmentation/grouping mechanism for approximate numerosity extraction from both real and ICs objects. Results are discussed in light of their significance for artificial intelligence models of visual perception.