A General-purpose Vision Processor with 160x80 Pixel-Parallel SIMD Processor Array

Provided by: Institute of Electrical & Electronic Engineers
Topic: Hardware
Format: PDF
Computer vision applications demand high computational power, due to the large amount of raw image data required to be processed at real-time speeds. Low- and medium-level image processing algorithms are characterized by a high degree of data parallelism and computational locality, i.e. every value in the result image is a function of the original pixel value and its bounded neighbourhood. In this paper, the authors present a vision processor, which incorporates a 160

Find By Topic