Parallelization of gradient-based edge detection algorithm on multicore processors

摘要

Current computers are multi-core, with more than one physical core in one microprocessor chip. Many applications in digital image processing are parallel in nature. Therefore, multi-core processors can be exploited to perform such computations in parallel. In this paper, the standard OpenMP threading library is used to speed-up the edge detection operation on multicore processors. Different partitioning methods of the input image are tested and their effect on the performance of the parallel implementation of the Sobel Edge Detection algorithm is analyzed. It is shown that the horizontal partitioning of the image leads to better performance than vertical partitioning or two-dimensional block partitioning. Various numbers of blocks of the image are tested. It is shown that a number of blocks equal to 0.25 the size of the cache line and a number of threads double the number of physical core give the best performance of the parallel Sobel algorithm.