Abstract
In visual crowding a target stimulus can be difficult to identify or discriminate when it is surrounded by additional flanking elements. Numerous empirical studies have demonstrated that the strength of crowding depends on the apparent grouping relationship between the target and flanking elements. Last year (Francis, Manassi & Herzog, 2016), we described a neural network model of visual perception that explained a wide variety of crowding effects as the result of neural grouping and segmentation mechanisms. We now present new model simulations of an uncrowding effect, where a single stimulus around a target causes strong crowding but additional flanking stimuli produce weaker crowding effects. Manassi, Lonchampt, Clarke & Herzog (JOV, 2016) investigated more than a dozen such uncrowding examples; and our new simulations demonstrate that the neural network model accounts for many of these new uncrowding examples because adding flanking elements leads to a larger perceptual group of flankers that is distinct from the target. This larger group is more easily segmented by top-down signals that produce distinct representations of the target and the group of flankers. These distinct representations avoid crowding effects because the segmentation process effectively isolates the target. Consistent with the new empirical data, the model demonstrates uncrowding for flankers made of squares, circles, hexagons, octagons, stars, and irregular shapes, as long as they form perceptual groups. The model fails to match empirical findings for some situations where grouping effects seem to be very sensitive to contextual details; a failure that is not too surprising since the model grouping mechanisms do not perfectly match human behavior. Overall, the model is able to account for many new cases of uncrowding; and the cases where the model is unsatisfactory suggest ways to improve the model to better understand crowding effects, perceptual grouping, and visual segmentation.
Meeting abstract presented at VSS 2017