The Connected Component Algorithm on The NON-VON Supercomputer

The NON-VON Supercomputer is a highly parallel tree-structured computer that is being Implemented at Columbia University. In this paper, we demonstrate that tree architectures with their favorable characteristics for VLSI Implementation, and fast global broadcast, lend themselves easily and naturally to the representation and manipulation of Images represented by hierarchical data structures A description of NON-VON architecture IS presented With an emphasis on the special architectural features that will be used m our Image understanding algorithms. We adopt a variation of the quad tree data structure, called the binary Image tree, to represent images in the NON-VON tree We show how Images are loaded in the NON-VON tree, and present the algorithm for budding the binary Image trees. An efficient Implementation of the connected component labeling algorithm on NON-VON is then presented Simulation results are discussed, and we show the fast execution time of the algorithm on NON-VON. Other algorithms are also developed, such as hlstogrammlng, Hough transform, Set operations and Image correlation, and we can conclude that NON-VON can be used to Implement efficiently several :important Image understanding task