ABSTRACT
The classic concept of recognition usually involved a procedure of revealing local signs followed by their analysis, using statistical or cluster heuristic methods. In the early years of research, experts working on recognition programs first tried to normalize an image, that is, to reduce it to a given size, orientation, and so forth-and then checked how it was similar to samples from a certain set [135). However, when recognizing objects, humans usc a preliminary information and can, in a number of problems, disregard changes in a figure's size and orientation, gaps in its image, and so forth. They do not need a completely normalized and noise-free image since, perceiving it as noise-distorted, incomplete, large or small, they arc capable of recognizing it.
