As revealed by the Gestalt school in the first half of the century, visual perception is governed by certain simple rules which group parts into wholes in accordance to ‘laws’ like grouping by proximity, similarity, closure, symmetry, and good continuation. Although these principles can be investigated by experiment, their underlying neural computation is largely unknown. It has been speculated that synchronisations of visual cortical neurons may serve as the carrier for the observed perceptual grouping phenomenon. We present a neural network for preattentive perceptual grouping derived from neurophysiological and psychophysical findings, incorporating a relaxation phase labeling and diffusion process. The network groups visual features into perceptual entities by (de)synchronising parametric phase labels of simple neural oscillators using a constraint satisfaction mechanism. The local constraints between features, which model the Gestaltist grouping principles of proximity and good continuation, act horizontally in and vertically between feature dimensions to allow for the emergent segregation of globally salient contours in phase space, suppressing false responses generated from the edge detection stage. By applying the grouping mechanism to various contour types ranging from dotted lines to intensity edges we show that the phase-based object representation is able to account for various perceptual phenomena like the closing of small contour gaps and the perception of illusory contours. Based solely on edge responses and local interactions thereupon, the neural dynamics allows the emergent formation of globally distinguishable objects in phase space, which can be extracted by an attentional mechanism tracking the spatially modulated phase information.
Neural dynamics of perceptual grouping: Textures, boundaries, and emergent segmentations
