Title :
Implementing the boundary contour system on a multi vector processor iPSC/2 HyperCube
Author :
Diamond, Andrew ; Holden, David
Author_Institution :
Gen. Dynamics, San Diego, CA, USA
Abstract :
The boundary contour system (BCS), a massively parallel neural network model designed to perform early vision tasks, is discussed. Its tasks include boundary detection and segmentation. An arrangement of short-range competition and long-range cooperation among orientation-sensitive cells ensures a rapid, active sharpening of boundaries, including emergent boundaries (boundaries between regions which do not differ in global statistics of luminance distributions or have luminance patterns that are discontinuous or fuzzy). Patterns of activity in the BCS restrict the diffusion of activity in a complementary system called the feature contour system, which fills in features that are delineated by the BCS. The BCS is a dynamical system, not an algorithm. It is defined by an architecture of neurons and a set of differential equations that govern the neurons. Its implementation on an Intel iPSC/2 Hypercube with 16 vector processor nodes is discussed, as are the hardware used, integration techniques, data structures, memory management techniques, synchronization schemes, and other implementation issues.<>
Keywords :
computerised pattern recognition; computerised picture processing; neural nets; parallel processing; boundary contour system; boundary detection; data structures; early vision tasks; edge detection; emergent boundaries; iPSC/2 HyperCube; integration techniques; long-range cooperation; luminance distributions; massively parallel neural network model; memory management techniques; multi vector processor; orientation-sensitive cells; pattern recognition; picture processing; segmentation; short-range competition; synchronization schemes; Image processing; Neural networks; Parallel processing; Pattern recognition;
Conference_Titel :
Neural Networks, 1989. IJCNN., International Joint Conference on
Conference_Location :
Washington, DC, USA
DOI :
10.1109/IJCNN.1989.118571