Approximation of class regions by decision trees and its application to handwriting recognition

Decision rules are key technology in the fields of decision making, data mining, knowledge discovery in databases, and so on. This paper introduces a new application field of decision rules, handwriting pattern classification. By applying decision rules into pattern recognition, one rule forms a hyper-rectangle in the feature space, i.e., every decision rule corresponds to one hyper-rectangle in the feature space. This means that a set of decision rules can be considered as a classification system, and this kind of classification system is called a subclass method. This paper applies decision rules to handwritten Japanese Kanji recognition. Then, it shows some of computer experimental results.