Fast Prototyping of Image Processing Applications Using Functional Skeletons on a MIMD-DM Architecture

development of an application performing connected component labelling on digital video streams. In this paper, we assess the applicability of the skeleton-based approach to portable parallel pro2 ~ l ~ ~ ~ i t h ~ i ~ skele ons gramming within the vision application domain. Four algorithmic skeletons for low and intermediate level image processing are proposed. For each skeleton we give an architecture-independent executable specification, a parallel implementation template as a process network and a performance model. These skeletons are used to build a programming environment dedicated to the fast prototyping of embedded vision applications, and mostly built from existing software components. Several examples are provided to illustrate the concepts and tools introduced, including a real-time vehicle detection and tracking application.