Still trying to drum up more interest in its Clipper RISC, Intergraph Corp’s Advanced Processor Division in Palo Alto, California is highlighting the fact that Philips Systems & Sondertechnik GmbH in Bremen, West Germany has designed and delivered an image processing system that uses lots of Clippers. The system, code named Samba, incorporates 20 of the microprocessors in a highly parallel system architecture that is loaded with pattern recognition software claimed to be capable of detecting and symbolically describing objects in near-real-time – symbolic computation is the analysis of objects represented as geometric shapes or luminance features rather than as clusters of pixels. According to Intergraph, symbolic computation has hitherto usually been done sequentially so that near-real-time object recognition has not been feasible – but if that shape looming up might be a sheep or a bear, it’s kind of important to be able to decide which it is in very near real time. Earlier simulations using a DEC VAX-11/750 – a machine that hardly offers impressive performance by today’s standards required about 15 minutes to process a single image, or eight full days to complete a typical 30-second image sequence. The parallel Clipper-based system reduces the processing time to 280mS per image, or three to four minutes for the entire image sequence – a 3,000-fold improvement. The system comprises two primary subsystems – specialised front-end video image-processing hardware, and the 20-Clipper processor core, and operation consists of three stages: object generation, object selection and reporting of results. After completing these computations, the system initiates the object selection phase, where objects that meet a predetermined set of features are picked. In this phase, the 20 Clippers operate in a tightly coupled mode as a mailbox chain of 10 master-slave pairs.
