Poster Presentation

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Local Texture Statistics are Sufficient for Category-level Contingent Capture

Yuhan Shi¹, Ryan S Williams¹, Susanne Ferber¹, Margaret M Henderson¹; ¹Carnegie Mellon University

Presenter: Yuhan Shi

Category-level contingent capture occurs when task-irrelevant distractors sharing the target's category impair target detection, yet the representational basis of this effect remains unclear. As prior studies have typically employed intact object images, which preserve both mid-level local texture statistics and global spatial configuration, it is unknown whether capture is driven by local texture statistics, global spatial configuration, or both. The present study used a rapid serial visual presentation (RSVP) paradigm to adjudicate between these possibilities. On each trial, participants (N = 74) were cued with a basic-level category and searched a central RSVP stream for a matching target. Two frames before the target, a parafoveal distractor appeared that either matched or mismatched the cued category and was either intact or scrambled using CNN-based texture synthesis (Gatys et al., 2015), preserving mid-level local texture statistics while disrupting global spatial configuration. Signal detection analyses showed that category-matching distractors reduced perceptual sensitivity (d') relative to mismatching distractors for both intact and scrambled images, with intact distractors resulting in a larger sensitivity cost relative to scrambled. These findings show that mid-level local texture statistics alone carry sufficient categorical information to elicit contingent capture in the absence of coherent object shape, but that global spatial configuration conveys additional information that further strengthens the attentional template.

Topic Area: Computational Models of Vision & Visual Cortex