Working towards efficient parallel computing of integral images on multi-core processors

This paper presents a novel multi-threaded parallel algorithm for computing integral images on multi-core processors. At each stage of the design, we evaluated several approaches, well-established as well as newly proposed. According to the results of the evaluations and our analysis the best suitable solutions have been identified, from which the parallel algorithm was synthesised. Tests were made and showed that on systems with fast FSB, e.g., 1333 MHz, when running with two threads bound on distinctive processors sharing the same L2 cache the implementation of the algorithm could run at a speed twice as fast as that of the best known sequential reference. Tests also revealed that L2 cache size, system bus speed, micro-architecture and topology of the processor all had their share in determining the performance of the implementation relative to the sequential reference.