결정트리 분류기를 이용한 무제약 필기체 영문자 오프라인 인식

초록

This paper proposes an algorithm for generating a decision tree classifier to recognize unconstrained handwritten characters effectively. ID3 learning algorithm is adoptied to construct the decision tree classifier automatically. At each internal node of the decision tree classifier, an optimal feature of a feature vector was selected according to the ID3 learning algorithm to separate training data to several branches. Consequently, the decision tree classifier had minimum size. Differenct characters may have same features. It is called feature conflict. To resolve the feature conflicts, recognition consists of two phases: decision-tree search and heuristic recognition. The heuristic recognition conducts for the resolution of the feature conflicts. As a feature, ESP(Extended Stroke Primitive) is considered and additional features are considered for the heuristic recognition. The experiments conducted in this study uses the ETL-1 database of unconstrained handwritten characters. Experimental results confirm that the proposed decision tree classifier is quite effective for the recognition of unconstrained handwritten characters.