Abstract
In crowding, perception of a target deteriorates when neighbored by flankers. Contrary to predictions of classic pooling models, crowding is strong only when the target groups with the flankers. We recently showed that a version of the neural LAMINART model could explain many grouping effects in crowding. In the model, top-down segmentation signals promote separate neural representations of separate groups in a scene. The model is implemented as spiking neurons in the NEST simulator, consists of hundreds of thousands of neurons and several million connections, and represents early stages of vision (V1, V2 and V4). Here, we present new simulations that hypothesize that the placement of the top-down signal is less precise for more peripheral locations. The overall strength of crowding for flankers depends on whether the top-down signal can generate distinct segmentations of the target and a flanker. A flanking set of 8 long lines spans a large surface that the top-down signal will easily catch and segment from the target vernier, so such segmentations will be very common regardless of distance from fixation. In contrast, a flanking square can, in principle, be segmented from the target, but such segmentations will be less common with larger distances from fixation. These properties produce the predicted crowding asymmetry: when the target is in the right-side visual field, crowding is stronger with long-sized lines flanking the target on the left (closer to fixation) and a square flanking the target on the right (farther from fixation) than when the flanker locations are switched. In an empirical study, 6 observers discriminated a target vernier with the stimuli used in the simulations. Consistent with the model predictions, crowding was stronger with an array of aligned flankers to the left and a square to the right compared to the other way around.
Meeting abstract presented at VSS 2017