Abstract
Previous empirical studies have demonstrated that crowding effects, where a visual target can be difficult to classify when surrounded by flanking elements on either side, are strongly altered by perceptual grouping of the target and flankers. Here, we describe a real-time neural model of perceptual grouping and segmentation that allows non-specific top-down signals to alter the representation of visual percepts. This top-down control allows an observer to generate separate representations of target and flanking elements in some visual crowding situations. Simulations show that the model properly accounts for many empirically measured grouping effects in crowding; namely a target is strongly crowded if it groups with the flankers but is hardly crowded at all if the target seems to be part of a distinct group. The model segmentation process explains why adding flanking elements can (otherwise paradoxically) reduce crowding and why seemingly tiny changes to the flankers can alter perceptual grouping and dramatically alter the effects of crowding. It also explains why target-flanker similarity produces the strongest crowding effects. These attention-like mechanisms explain how observers interact with the visual representation of a scene to enable them to solve specific perceptual tasks
Meeting abstract presented at VSS 2016