Representation of objects and object features in the primate visual temporal cortex: new evidence and insights.

5th ICN La Jatta UCSD, August, 23-28, 1998

     The visual system in our brain must be equipped with a mechanism by which we are able to: (1) recognize numerous objects in the world, (2) identify objects irrespective of changes in illumination, viewing distances and angles, and (3) reconstruct 3-D structures of objects from 2-D retinal images.  Studies in monkeys have found the possible neural correlates of the mechanism.  Most neurons in area TE of the inferior temporal cortex, the final stage of the object vision pathway in the macaque monkey, respond to complex object features such as shape or shape combined with color or texture.  The features critical for activating TE neurons, are, however, not complex enough to represent a natural object by the activity of a single neuron.  Most TE neurons are also selective for a particular orientation of the preferred features.  It is thus generally assumed that the activity of a group of neurons is necessary to represent an object.  Neurons with similar stimulus preferences are clustered in columns roughly 0.5mm in width.  Evidence for the columnar organization of these neurons has been obtained from single-neuron recordings and optical imaging studies in anesthetized monkeys, and from multiple-neuron recordings in awake monkeys.  An emerging idea is that the visual features encoded by the activity of columns constitute gvisual alphabetsh which in combination allow representatopm of an object.  This combination scheme contrasts with the egnostic cellf hypothesis as well as the purely distributed coding scheme.  Our recent study shows that TE neurons are not only selective for 2-D visual features, but are also sensitive to binocular disparity.  Moreover, some of them change their activity depending on the perceived surface structure regardless of the type of disparity added to the figure.  We suggest that the object vision pathway, including area TE, is important for reconstructing 3-D surface structures of objects.

(Supported: CREST, MSSEC)