Principles of Global Visual Processing of Local Features can be Investigated with Parallel Single-Cell- and Group-Recordings from the Visual Cortex

Abstract

In order to investigate cortical mechanisms of visual integration, we analyzed relations between feature coding by single neurons and groups of neurons in cat visual cortex (Eckhorn et al. 1988). Local visual coding in the primary cortical areas is well understood and can be characterized by the receptive field (RF) properties of single neurons. RF-properties are due to relatively invariant (though nonlinear) spatio-temporal filter properties of the network between receptors and the neuron in question. Neurons with similar RFs are retino-topically organized in a number of interconnected “feature maps” (visual cortical areas; Hubel and Wiesel 1962, 1963). The constant relations between retinal surface, retinotopic cortical maps and RF-properties probably form invariant frames of reference between outer visual world and perception. However, in order to cope with the immense variety of local feature combinations in natural scenes, flexible mechanisms of feature association are required. To date flexible principles of global visual processing are barely understood and there are no proven theories of how local coding properties of single cortical neurons are flexibly linked into a coherent global percept.