A machine independent approach to parallel algorithm design on image computation

A unified parallel algorithm based on the stripcode image data structure of Arabnia et al. (1986) is proposed for image rotation. The approach adopted represents a middle path between classical multiprogramming techniques, i.e. architectural driven machine models and a variety of programming paradigms (e.g. functional programming and object-oriented programming). The implementation is designed by introducing algorithmic parallelism to structure a computational model while not losing touch with the intrinsic parallelism available in the hardware. The algorithm is targeted for the AMT DAP 510 (single-instruction multiple-data machine) and an array of transputers (multiple-instruction multiple-data machine). The algorithm performs the rotation of a digitized image through an arbitrary angle 0<Θ<π/2. Implementations of this algorithm on the above hardware are described and timings of these implementations are presented to compare with previous work