Parallel implementation for image rotation using parallel virtual machine (PVM)

Many image processing algorithms are particularly suited to distributed computing because these images are difficult and time consuming to analyse. Furthermore, existing algorithms contain explicit parallelism, which can be efficiently exploited by processing arrays. A good example of an image processing operation is the geometric rotation of a rectangular bitmap. This paper shows how this can be implemented on a distributed system using a parallel virtual machine, by splitting images into number of parts and sending each to a separate computing node. Each node performs a rotation on its partial image before returning it to the master node to be recombined in a single image. A variety of image sizes and number of distributed computing nodes were used to determine the efficiency of this technique, and whether it offers enough speed improvement to justify its complexity. Whilst rotating large images benefited enormously using this algorithm, small images rotated more slowly than they would have done on a single processor. This is of particular importance in the case of large digital images, which may consist of millions of pixels.