Visual Perception of Depth-from-Occlusion: A Neural Network
PENNSYLVANIA UNIV PHILADELPHIA DEPT OF BIOENGINEERING
Pagination or Media Count:
Two major goals have been accomplished during our first year of funded research. First, we have developed an environment for simulating neural systems, called NEXUS. NEXUS is designed for studying large networks. Towards this end, it is based on the principles of topological map organization, and introduces a novel network construct, programmable generalized neural PGN units. A single PGN unit can emulate the behavior of an entire neural circuit or assembly, allowing complex systems to be simulated. NEXUS is window-based and features an intuitive graphical user interface. Our second goal has been the development of a model of how the cortex extracts depth-from-occlusion. The model utilizes a multidistributed representation of depth and emphasizes how object segregation and discrimination occur. The model shows how spatially separated portions of an occluded object can be dynamically linked in mental representations e.g. the moon viewed through tree branches is perceived as an intact circular disc, not as separate pieces. Early tests also indicate that the model will fully account for the variations in the vividness of perception of a wide range of illusory Kanizsa contours. The model is currently being simulated and tested using NEXUS.