MASSIVELY-PARALLEL DATA-FLOW COMPUTER DEDICATED TO REAL-TIME IMAGE-PROCESSING

This paper describes a massively parallel dataflow computer dedicated to real-time image processing. Two types of data-driven processing ele ments are embedded into a regular cubic mesh: 1024 low level custom da taflow processors (50Mops/s each) and 36 T800 transputers. The program ming environment relies on an extensive use of the functional programm ing (FP) concept. An image processing algorithm first expressed in an FP-like functional syntax is converted into a dataflow graph, the node s of which are realtime operators to be implemented on physical proces sors by direct mapping onto the array. The concepts and the architectu re, both hardware and software, are explained in detail. We conclude w ith an example of a complete implementation: a connected components la beling working in real time. (C) 1997 John Wiley & Sons, Inc.