A real-time parallel image-processing model

Image processing can be viewed as a data-reduction pyramid with pixel-based, low-level, processing at its base and semantically-based, high-level, processing at the apex. It is the base of the pyramid which represents the bottleneck. However, it is unclear whether the parallel computation models so far proposed address the needs of low-level image processing. This paper therefore proposes a real-time parallel processing model for image processing. A fast Fourier transform (FFT) for batch-processing of images illustrates the model in a distributed workstation environment. The model has also been implemented on a dedicated modular parallel machine, a Transtech Paramid, a C40 parallel DSP network, and the Unix-like real-time operating system VxWorks. The intention is to provide a common processing environment across accessible parallel architectures, including development support tools for the application design process from initial sequential simulation, through parallel decomposition to embedded parallel application