The partition-combination method for recognition of handwritten characters

The partition-combination method is a universal strategy for studying all sciences. This paper introduces such a strategy to handwritten character recognition, which is developed from our previous study. Let a pattern be split into sub-patterns, or called parts, bases, roots, etc. The easier part recognition is first carried out, then recognition of the entire pattern can be completed by integrating the results of part recognition. In this paper, the computational formulas for evaluating the recognition rates of parts and their combinations are derived, and a number of fascinating results have been reported. Many new combinations of parts have been found, leading to better recognition in practical applications. Numerical experiments have also been conducted using 89 patterns of the most frequently used alphanumeric handprints, leading to the discovery of several interesting aspects related to character recognition. Furthermore, human behavior in handwriting may be discovered based on the computational data obtained from the new algorithms. (C) 2000 Elsevier Science B.V. All rights reserved.