Pattern Detection and Recognition

Abstract

Two types of pattern-processing problems are discussed in this paper. The first, termed "pattern detection," consists of examining an arbitrary set of figures and selecting those having some specified form. The second problem, "pattern recognition," consists of identifying a given figure which is known to belong to one of a finite set of classes. This is the problem encountered when reading alphanumeric characters. Both recognition and detection have been successfully carried out on an IBM 704 computer which was programmed to simulate a spatial computer (a stored-program machine comprised of a master control unit directing a network of logical modules). One of the programs tested consisted of a recognition process for reading hand-lettered sans-serif alphanumeric characters. This process permits large variations in the size, shape, and proportions of the input figures and can tolerate random noise when it is well scattered in small specks. Programs for detecting L-shaped (or A-shaped) figures in the presence of other randomly drawn patterns have also been successfully tested.