A system for automatic recognition of handwritten words

Abstract

The recognition of handwriting can be considered an important problem in the general pattern recognition area because the set of patterns, say individual words, possesses a degree of variability that far exceeds that of problems where relatively good solutions have been previously found. Whereas in the case of character recognition the number of pattern classes considered different usually does not exceed one hundred, the number of pattern classes with which one finds himself confronted here is only limited by the vocabulary of the language. Furthermore, the problem is reasonably well-defined, i.e., in most cases the correct categorization choice is known by comparison with human performance. In some cases such performance by different people may not result in complete agreement, but even then the number of alternative results is restricted to a small number. Experimental data are readily available and their variability, insofar as they depend on subject and context, can be easily controlled.