Fast Hough Transform on Multiprocessors: A Branch and Bound Approach

The algorithm of the Fast Hough Transform presents a strong irregularity which makes its parallelization difficult, especially if we introduce certain improvements into the sequential algorithm, such as the elimination of straight lines. In this work we approach this algorithm as a branch and bound problem, developing three different parallel algorithms. The first of these algorithms is based on an initial distribution of the problem that guarantees a good balance of computations. The other two algorithms perform a dynamic distribution of the load using different policies. In one of them, the mechanism is distributed and rebalancing is carried out at the demand of the processors left without load. In the other, one processor acts as a central processor and decides when to carry out the rebalance and what the correspondence between overloaded and underloaded processors will be. © 1997 Academic Press