Mechanization of human ability to recognize the complex line patterns such as hand-printed charact ers and fingerprints has long been the common aim of engineers and biologists. The use of scanning techniques which compare a sample with a master template for acceptable discrimination between characters seems to be of little avail in the case of hand-printed character recognition owing to the lack of character invariants which make it possible to correctly recognize the various personalized writings. That is to say, it is quite free to let the hand-printing have such characteristics as size, location, slant, rotation and others which personalize the hand-printing.
This paper is concerned with the statistical method of hand-printed character recognition in a nume rical manner suitable for machine discrimination of hand-printed characters and similar line patterns. The operation rule of the categorizer is generally called the recognition function. The statistical recog nition function based upon a recognizability criterion for hand-printed characters has come from the introduction of a mathematical concept “distance” between the characters mapped into an
N-dimensional parameter space under certain assumptions about the probability distribution of measurements.
A practical example involving the statistical recognition of hand-printed characters I, J and K is solved by means of the above mentioned techniques. The recognition accuracy under a level of singnificance is also experimentally derived. By this statistical recognition method, it can be practically found out that how effective such a parameter mapping is and further the experimental results obtained in this paper are proved not to be eccentric at all.
抄録全体を表示