Adaptive algorithm and feature recognition system

An adaptive pattern-recognition algorithm is described. It has the followin g distinctive features: a compact description of the characteristic volumes of the alphabet objects in feature space is created; a training sample is synthesized from a single image of the object; the positions of reference p oints in feature space are determined by the size of the object being recog nized; a "fuzzy" decision rule with calculation of the confidence measures to hypotheses for assigning the object being recognized to one of the alpha bet objects is employed. A simpler version of the algorithm described is al so considered. Instead of employing a fuzzy decision rule, the object being recognized is assigned to the alphabet object to which the Euclidean dista nce in feature space is smallest. The results of the recognition of geometr ic figures on a noise-contaminated discrete image by these algorithms are c ompared with the results of recognition according the "nearest-neighbor" ru le and visual-recognition data. A feature recognition program system develo ped on the basis of the adaptive algorithm with a fuzzy decision rule is de scribed. (C) 2000 The Optical Society of America.