Top-down influence in early visual processing: a Bayesian perspective

Traditional views of visual processing suggest that early visual neurons are static spatiotemporal filters that extract local features by feed forward computation. The extracted information is then fed forward through a chain of modules to successively higher visual areas for further analysis. Recording from early visual neurons in awake behaving monkeys, we revealed there are many levels of complexity in the information processing of the early visual cortex. We found that the early visual neurons not only are sensitive to features within their receptive fields (RFs) but also to the global context of a visual scene, the behavioral relevance of the stimuli and the experience of the animals. These findings suggest that the early visual cortex (V1 and V2) is tightly coupled to and highly interactive with the rest of the visual system. The top-down interaction, mediated by recurrent feedback connections, introduces contextual information to influence the perceptual inference in the early visual cortex. (C) 2002 Elsevier Science Inc. All rights reserved.