Ed Connor Ph.D
Telephone Number: 410-516-7342
Fax Number: 410-516-8648
Johns Hopkins University, 3400 N. Charles St., Baltimore, MD 21218
Object Synthesis in Higher Level Visual Cortex
Our focus is on shape representation in the ventral, object-processing pathway of primate visual cortex, which includes areas V1 (primary visual cortex), V2, V4, and several further stages in inferior temporal cortex. This pathway functions to represent, interpret, store, and recall information about visual objects, and thus underlies much of our ability to recognize other individuals, interpret their facial expressions, comprehend symbolic written information, and generally to identify and interact appropriately with items in the environment. These impressive feats of visual information processing imply complex neural mechanisms that far surpass any artifical systems yet devised.
The nature of these neural mechanisms remains largely unknown. We are addressing this issue by studying how shape information is represented by large populations of cells at various stages in the ventral visual pathway. Our results show that shapes are represented in terms of their component parts. Individual cells in area V4 and inferotemporal cortex are explicitly tuned for boundary curvature, medial axis conformation, and relative position of shape components. Thus, for example, a given cell might respond to all shapes that contain downward-pointing convex projections near the right.
We characterize these tuning properties with mathematical functions in high-dimensional shape space. This allows us to combine signals across cells to reconstruct the population representation of any given shape (see Figure). Thus, we can "read out" the neural code for object shape from the response rates of neurons. While most cells appear to represent 2-dimensional shape information, we have also found that some cells represent 3-dimensional shape. In area V4, approximately 25% of shape-responsive neurons are tuned for the 3-D orientation of edges and lines in visual images. In the future we plan to study more complex aspects of 3-D shape representation.