Computational aspects of grouping explain visual crowding across space and time

In crowding, perception of a target deteriorates in the presence of flanking elements. Surprisingly, crowding can sometimes be ameliorated when more flanking elements are presented, a phenomenon called uncrowding. It was previously shown that uncrowding occurs when the target ungroups from the flankers. Here, we show that grouping processes can explain not only spatial interactions in crowding, but also interactions in the time domain. First, we show that grouping requires a minimum stimulus duration to occur: when participants discriminated the offset of a target vernier presented alone or flanked by lines or cuboids, a cuboid duration of at least 160 ms was needed for uncrowding to occur. Second, we show that the grouping process can be initiated by presenting only the cuboids for 20 ms before an ISI and then the display with the cuboids and the target vernier for 20 ms. With the preview, uncrowding occurs for short ISIs of 20 ms up to ISIs of 250 ms, pointing out to recurrent grouping processes taking place. Third, when presenting flanking elements during the ISI, the uncrowding effect occurred only when the elements formed a good Gestalt. We show that this body of results can be well explained by the Laminart model, in which recurrent processing segments the visual scene into different objects, which are represented in separate segmentation layers. When the target and flanking cuboids are processed in different layers, uncrowding occurs; when they are in the same layer, crowding occurs. Importantly, the preview of the cuboids gives the model sufficient time to segment the vernier target away from the cuboids. Taken together, our results highlight the importance of recurrent grouping processes in spatial and temporal interactions in vision.