The paper presents a data and task parallel low-level imageprocessing environment for distributed memory systems. Image processingoperators are parallelized by data decomposition using algorithmicskeletons. At the application level we use task decomposition, based onthe Image Application Task Graph. In this way, an image processingapplication can be parallelized both by data and task decomposition, andthus better speed-ups can be obtained. We validate our method on themulti-baseline stereo vision application
展开▼