Skeletonization of noisy handdrawn symbols using parallel operations

摘要

Digital images of handdrawn characters are treated by a noise insensitive process to yield a skeleton-like structure which preserves the main features of the original stroke pattern, but has a local syntax which allows its points to be classified in a way useful for recognition of the pattern.The computations, although lengthy, if executed in series, take about a millisecond on a proposed parallel picture processor with a retina like structure of two-dimensional words of a few hundred bits coupled to a 256 word 8 bit processor serving mainly to supply instructions