Abstract
Vision is often understood as a hierarchical, feedforward process, where visual processing proceeds from low-level features to high-level representations. Within tens of milliseconds, the fundamental features of the percept are established. Traditional models use this framework to explain visual crowding, where nearby elements impair target perception with minimal influence from stimulus duration. Here, we show that, at least for more complex displays, crowding involves highly dynamic processes. We determined vernier offset discrimination thresholds for different flanker configurations. In Experiment 1, for a 160-ms stimulus duration, crowding was lower for flanking Cubes/Rectangles compared to Lines, pointing toward underlying grouping processes. However, strong crowding occurred in all conditions at 20 ms, showing that grouping requires a minimum stimulus duration to occur. In Experiment 2, the crowded vernier (20 ms) was preceded by a 20-ms Cubes display. This brief preview led to uncrowding of the subsequently presented flanked vernier, but only for flankers that ungroup for longer durations (i.e., Cubes). This uncrowding effect occurred for time spans up to 1 s (Experiment 3) but could be interrupted by elements presented between the preview and the flanked vernier (Experiment 4). Our findings are well predicted by the LAMINART model, which employs recurrent segmentation processes unfolding over time to separate objects into distinct representation layers. Taken together, our novel preview effect highlights the importance of spatiotemporal grouping in crowding. In contrast to classic feedforward models, we propose that crowding is a dynamic process where multiple interpretations are modulated and gated by grouping mechanisms evolving over time.