Spatial context is known to affect how a stimulus is perceived. A variety of illusions spanning all visual modalities have been attributed to contextual processing, but their neural underpinning remains poorly understood. The goal of this thesis is to draft a computational theory of contextual processing across visual modalities that models neural populations in the early visual cortex. We begin by reviewing basic (i.e., non-contextual and feed-forward) cortical computations. We use the example of contour detection as a first illustration of how recurrent network models may bridge the gap from neurophysiology to behavior. Then, we present our main contribution: a novel recurrent network model of classical and extra-classical receptive fields, constrained by the physiology of the primary visual cortex. A key feature is the postulated existence of two spatially-disjoint (near vs. far) extra-classical regions with complementary contributions (facilitatory vs. suppressive) to the classical receptive field. The model accounts for a variety of contextual phenomena across visual modalities; in particular, it explains how center-surround interactions may shift from perceptual attraction to perceptual repulsion in tilt effects and from contrast to assimilation in induction effects. Finally, we discuss how the model reveals a novel taxonomy of contextual phenomena across visual modalities that is based on the different ways a stimulus may activate the extra-classical receptive field (exclusively, competitively, cooperatively) and on the nature of the stimulus (ambiguous vs. non-ambiguous). Overall, the ability of the model to account for the variety and complexity of neurophysiology and psychophysics phenomena provides computational evidence for a novel canonical circuit shared across visual modalities.
Mély, David Alex,
"Circuits and Mechanisms of Contextual Integration in Visual Cortex"
Cognitive Sciences Theses and Dissertations.
Brown Digital Repository. Brown University Library.