Summary form only given, as follows. The topic of multiprocessor computer architectures and parallel algorithms for computer vision is not new, but researchers are now addressing both a wider scope of issues and emphasizing system integration. Recently, a wide variety of new systems has been designed, built, and tested on a range of image analysis tasks. A critical question is how to achieve high performance in a complete, integrated set of component vision processes. A number of recent approaches to improving the performance of vision architectures are described. Comparisons are made relating the underlying model of parallel processing, the granularity of parallelism, and performance evaluation on various tasks covering several image representations and processing requirements.