Abstract
Stimuli outside classical receptive fields have been shown to influence significantly the activities of neurons in primary visual cortex. While there has been substantial experimental and theoretical interest in the phenomena and mechanisms of these contextual influences, computational insight
into their roles in visual processing is limited.
It is proposed that contextual influences are used for visual segmentation. The cortex locates GLOBAL region boundaries by detecting the breakdown of homogeneity or translation invariance in the input, using LOCAL intracortical interactions mediated by horizontal connections. This is implemented in a biologically based model of V1, in which contextual influences highlight neural activities near input region boundaries. Being more salient, the boundaries can pop-out perceptually for segmentation, as demonstrated in texture segmentation and figure ^ ground segregation. Our proposal introduces a new framework for visual segmentationˆsegmentation without
classificationˆnamely, segmentation occurs without classification of features within a region or comparison of features between regions. This framework is simpler than traditional approaches, making it implementable through V1 mechanisms. However, it is powerful enough to handle
segmentation problems that are tricky for traditional approaches. The model is suggested as performing pre-attentive segmentation; higher-level mechanisms are needed to refine its output.
In addition, the model performs segmentation in exactly the same neural circuit that solves the
dual problem of the enhancement of contours (and displays the usual phenomena of contextual
influences such as iso-orientation suppression and its contrast dependence). Theoretical predictions are discussed to propose experimental tests of the theory.
