Average orientation perception exploits mid-level grouping cues

Poster Presentation 33.315: Sunday, May 17, 2026, 8:30 am – 12:30 pm, Banyan Breezeway
Session: Perceptual Organization: Ensembles

Elijah Schwartz1 (), Gwen Holley1, Jason Haberman1; 1Rhodes College

Humans readily perceive the summary statistics from an ensemble, whether composed of oriented gabor patches or faces varying in expression. A pervasive question in ensemble perception research centers on understanding the hierarchical level at which statistical representation occurs. Seminal work suggests multiple levels of ensemble representation (e.g., Haberman, Brady, and Alvarez, 2015), but it is unknown whether the level of representation for a particular ensemble may be influenced by the strength of perceptual grouping cues. Here, we leveraged mid-level perceptual structures (i.e., Kanizsa figures/illusory contours) to explore whether implicit grouping cues might confer an advantage in average orientation judgments. Participants viewed either individual Kanizsa triangles or ensembles of four Kanizsa triangles constructed using traditional ‘pac-man’ shapes (i.e., inducers). These figures are powerful grouping cues, drawing upon dedicated neural mechanisms within area V2 and beyond. When the inducers are rotated, the vertices of the triangles remain but the illusory contours perceptually vanish. We asked participants to adjust a test triangle to match the orientation of the individual or the average orientation of the ensemble, when the inducers were rotated to either give rise to illusory contours or not. Overall performance was significantly worse when illusory contours were not visible. Most striking, however, was the significant interaction between stimulus type (Kanizsa or no Kanizsa) and set type (individual or ensemble) — participants were significantly worse in perceiving the average orientation of ensembles with no illusory contours compared to the orientation of a single triangle with no illusory contours. These results suggest ensemble perception efficiently leverages grouping cues to generate a more precise average representation, and also point to a higher-level of representation than might be expected, given that illusory contours do not emerge until extrastriate regions.

Acknowledgements: Rhodes College