Abstract
Gestalt cues are thought to organize visual input into meaningful perceptual units during pre-attentive stages of human visual information processing. As these units then serve as input for further perceptual and cognitive processing, characterizing the influence of Gestalt cues on their formation can lead to a better understanding of how the visual system initially organizes sensory input, as well as to the development of techniques to better present observers with critical visual information. Towards these ends, we used ERPs to a segmentation task in conjunction with psychophysical measures to map the time courses of several Gestalt cues. Observers determined whether patterns of 100 Gabor patches formed rows or columns. Patterns were formed by parametrically varying the tilt, spatial frequency, contrast, length, separation, and backgrounds of the patches. Overall, Gestalt versus Uniform stimuli showed faster correct response times, modulated P3 amplitudes, and attenuated P1 amplitudes. Relative to Uniform stimuli, stimuli correctly grouped by Proximity and Connectedness generally showed longer correct response times, later and smaller P3s, sooner and smaller P1s, and smaller N2s than stimuli grouped by Similarity of Spatial Frequency or Contrast. In contrast, stimuli grouped by Common Region and Similarity of Tilt generally showed shorter correct response times, sooner and larger P3s, later and larger P1s, and larger N2s. We next use a machine-learning algorithm to decode the characteristic ERP signatures of each Gestalt heuristic. In turn, we use these signatures to predict how and when an observer will parse a novel visual stimulus based on cortical activity specific to its Gestalt context, as well as when, and to what extent several visual illusions, such as the Ponzo, Müller-Lyer, Rod-and-Frame, and Simultaneous Tilt Illusions (reflecting perceived depth, size, and orientation) are affected by each type of Gestalt cue.