Development of an image processing system for texture segmentation

The development of a two-level machine vision system for the fast segmentation of textured images. A transputer network was used to realize the upper level, thus providing a powerful, expandable computing platform on which to execute intermediate and high-level vision algorithms. Low-level vision operations were delegated to a custom-developed pipeline card, permitting cost-effective, in-process execution of the required texture segmentation functions. The use of alternative data scans on the pipeline card further reduced hardware costs by allowing the decomposition of large moving average filters into one-dimensional filters performed orthogonally on the data set. This required fewer image processing ICs.<
>