Did you see the sound? A Bayesian Perspective on Crossmodal Perception in Low Vision

Abstract

Multisensory integration is often assumed to increase when visual input is degraded, yet it re-mains unclear whether low vision enhances susceptibility to cross-modal illusions or whether such effects depend on local variations in visual reliability. We tested low vision and sighted control participants on the Double Flash Illusion across 24 visual-field locations while sepa-rately measuring flash-detection accuracy. Both groups showed the expected auditory-driven increase in perceived flash numerosity, but only sighted controls reliably experienced the clas-sic "double-flash" percept. Illusion strength did not vary with eccentricity; instead, it was strongly predicted by local flash detection accuracy, indicating that sound-induced percepts depend on the availability of a reliable visual signal. Bayesian Causal Inference modeling revealed substantially weaker and more variable fits for low vision observers, with poorer fits associated with reduced visual sensitivity. Although model parameters did not differ significantly between groups, the similarity in estimated visual noise likely reflects model limitations rather than true equivalence in sensory precision. Together, these findings show that low vision does not globally amplify audiovisual interactions; rather, auditory enhance-ment depends on local visual reliability, and degraded vision leads to weaker alignment with Bayesian-optimal predictions.