手書き数字の統計的解析 : 3つのレベルの主成分分析による文字クラスのプロトタイプの決定

概要

論文の詳細を見る
The aim of this paper is to determine the set of substroke classes of discrete handwritten numerics. From this set a prototype of any letter class can be constructed. In this paper, on-line discrete handwritten characters are used for a basic data for statistical analyses. Firstly a real handwritten character on a tablet digitizer is converted to a time series of two-dimensional points. Then the time series is approximated in the form of a sequence of the inclination angles of n short equal-length line segments. The sequence of the inclination angles is represented by a n-dimensional vector. Secondly, three levels of the principal component analyses are made for the set of n-dimensional vectors representing discrete handwritten characters in the numerical letter classes. These three levels of the principal component analyses are : that for a single letter class, that for a pair of letter classes, and that for all given letter classes respectively. In determining the set of the substroke classes, the concept of the functional distinctive feature by which ambiguous characters can be discriminated is proposed.
東海大学の論文