Abstract
The Hough Transform (HT) is known to be a powerful technique in shape recognition and motion analysis. On shared-memory multi-processors, Image Partitioning and Parameter Partitioning are data partitioning techniques which give rise to two different classes of MIMD algorithms for HT. These techniques differ in terms of data locality and task granularity. In this paper, we compare these algorithms by the trade-offs involved in their mapping on bus-based shared memory machines. Based on our analysis, we suggest an efficient implementation of Parameter Partitioning which improves on known results. The improved performance is reflected in the execution times obtained on a Sequent Balance machine. Our analysis is also verified by running the algorithms on Proteus, a multi-processor simulator. The techniques of Image Partitioning and Parameter Partitioning are then extended to hypercube multiprocessors. It is shown that Parameter Partitioning performs better on hypercubes also. The measurements on hypercube are obtained by running the algorithms on a 64-node nCube machine.
Datta, Abhijit (1994). Efficient implementation of hough transform on multiprocessors. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -1994 -THESIS -D234.