利用总体特征对中文签字进行分类的聚类算法

Dynamic clustering algorithm classifying same Chinese character based on global features

针对不同书写者书写同一字的分类问题,在C-均值法和马氏距离测度的基础之上,提出了一种动态聚类算法,并讨论了签字的总体特征选择问题。利用该聚类算法对不同书写者的签字进行二分分类得到了较好的效果。实验显示,选择一组代表书写者书写风格的特征是分类成败的关键。文中选取的五个总体特征应用到非模仿的签字鉴别中有较好效果。

To classify the same Chinese characters written by different writers, a dynamic clustering algorithm was presented in this paper. The algorithm was based on C-means and Mahalanobis distance. Firstly, the patterns were classified using C-means based Euclidean distance. Secondly, according to the value of a principal function, the class of each pattern was adjusted. At last, based on the initial classes, all patterns were classified again based on Mahalanobis distance. Except the clustering algorithm, the selection of character features was discussed too. Experiments result showed it is very important to select a set of features that represent accurately the handwritten style of writer. A promising result has been obtained by using this algorithm to classify same characters written by different writers.