A two-step strategy for character recognition using geometrical moments

This paper discusses the use of geometrical moments as possible information-extracting features for recognizing characters. The development of decision criteria is traced from the normal multivariate probability function to the Hamming distance, and the relative efficiencies compared. Decision boundaries are investigated by a stepwise transformation from character to character. Computer simulation of one-step recognition processes using these discriminants yields recognition rates which vary from low values using the simplest criterion, up to 100% in some cases using the most complicated criterion. Combinations of these methods into two-step processes are compared, and an attempt is made to find an optimum strategy for the case of poor quality hand-drawn capital letters.