PATTERN RECOGNITION USING AN ASSOCIATIVE STORE

1,230,834. Character recognition; associative stores. INTERNATIONAL BUSINESS MACHINES CORP. 23 Oct., 1968, No. 50202/68. Headings G4C and G4R. A set of d binary symmetric (i.e. possible values + 1 and -1) numbers, e.g. from matrix optical sensing of a character to be recognized, is classified by using it to interrogate entries in an associative store representing the coefficients, restricted to possible values + 1, 0, -1, of the equations of planes in d-space which divide the space into volumes such that substantially all points in the same volume represent sets of numbers belonging to the same class, to obtain an indication of on which side of each plane the point representing the set of numbers to be classified lies, these indications being used to interrogate other entries in the same (or a different) associative store to identify the volume in which the point lies. The store holds two words for each plane, holding the coefficients in true and complement form respectively, the complements of + 1, 0, -1 being -1, 0, + 1 respectively, and + 1, -1, 0 being stored as 1, 0 and " don't care " respectively. " Don't care " gives a match whatever the interrogating input. Analogue voltages, one from each word, represent the extent of mismatch, and these two voltages for each plane are subtracted in a differential amplifier to set a latch or not according to the sign of the result to indicate on which side of the plane the point lies. Alternatively, interrogation for a given plane may be serial by word and bit each bit mismatch with one word incrementing a counter and each with the other decrementing it. The " which side of plane " bits, set into latches, are used to interrogate the second set of entries, each of which includes a set of such bits for a respective volume and a name associated with the volume which is read out to identify the character.