Abstract

Utilizing large amounts of data for repetitive tasks, machine vision falls under the category of processes requiring very fast limited instruction set computers. This paper describes the Parallel Micro Image Processor (PMIP), a general purpose SIMD 256-processor micro-sequencer controlled image processor. Danielsson and Levialdi describe a method for computing the "degree of parallelism" 1 which tells the speed potential of a system. The PMIP has a degree of parallelism of 2560 compared with 112 for GLOPR (Perkin Elmer) and 6336 for the Cytocomputer (ERIM). Section 2 describes the host interface. Section 3 describes the video I/O structure of the PMIP. Of particular interest is its high bandwidth and variable offset grabbing capability within the field-of-view (FOV). Section 4 describes the processor architecture itself, section 5 describes the specialized feature extraction hardware and section 6 describes the programming environment.

© 1987 Optical Society of America