Crowding asymmetries explained by a model of image segmentation

In crowding, perception of a target deteriorates when neighboring flankers are presented. Flankers close to the fovea deteriorate performance less strongly than flankers presented peripherally, the well-known crowding asymmetries. For example, we presented a vernier at 4 degrees of eccentricity in the right visual field and observer discriminated its offset direction. Performance was better when the vernier was flanked by a square on the left and 8 long lines on the right than the other way around. We recently showed that a version of the LAMINART model explains many effects in crowding well. The model comprises hundreds of thousands of spiking neurons mimicking early stages of vision (V1, V2 and V4). In addition, a segmentation network parses the visual stimulus in different segmentation layers. The segmentation process is started by a top-down segmentation signal. The precision of cortical segmentation signals follows cortical magnification and thus decreases with eccentricity. The model could well reproduce the results of the experiment described above, because the representation of the flankers close to the fovea is denser, which facilitates segmentation. For both conditions, the 8 lines span a large area that is easily caught by segmentation signals, independently of their position in the visual field. In contrast, the smaller square is more difficult to catch in the periphery than close to the fovea.