Abstract
Disconnected elements in a visual scene are often interpreted as forming groups or objects. While there are hypothesized grouping rules, such as co-linearity, smoothness, or similarity, that predict some types of grouping, there are many situations where these rules do not seem to apply. In particular, it is possible for an observer to treat a single element as a distinct stimulus but, with a different set of instructions, to treat the element as part of a group. In contrast, neural models of perceptual grouping establish the range of grouping by the spatial size of connections: quite disparate elements can (and must) form a group only if there exist long-range connections. An example is the LAMINART model, which uses properties of grouping to explain a variety of perceptual and neurophysiological phenomena. Here, we introduce a grouping circuit that uses only local connections. Like the bipole cells of the LAMINART model, the new circuit forms illusory contours that connect elements of a group by requiring sufficient input from at least two visual elements. Critically, the new circuit trades off space and time: excitatory signals spread outward until a damping signal arrives to inhibit neural activity that is not supported by at least two sources of excitation. Increasing the delay between excitatory spreading and inhibitory damping allows illusory contours to form over longer distances. We hypothesize that perceptual grouping is subject to top-down control by modifying this relative delay, and that observers can use such top-down control to help perform different types of tasks, such as selecting a group of elements or focusing on a single isolated element in a visual scene. We describe the circuit and show simulations of how it can be applied to various situations
Acknowledgement: Horizon 2020 Framework Programme (Grant Agreement No. 785907 (Human Brain Project SGA2))