A hybrid SIMD/MIMD architecture for image understanding

This paper presents an integrated parallel architecture for image analysis and computer vision. The architecture consists of an SIMD array processor with multiple broadcast busses, an MIMD multiprocessor with orthogonal access busses, and a two-dimensional memory array shared by both processor systems. Low-level image processing is performed on the SIMD array, while intermediate and high-level image analysis is performed on the multiprocessor system. The interaction between the two levels is supported by a common shared memory. To illustrate the power of such a two-level system we show that a wide class of image and vision analysis tasks can be performed efficiently on the architecture. Representative problems include transitive closure, histogramming image component labeling, pyramid computation, Hough transform, pattern clustering, and scene labeling.