基于3种测度值的特征提取方法优化评价.pdfVIP

学兔兔 第31卷 第4期 仪 器 仪 表 学 报 V01．3l No．4 2010年4月 Chinese Journal of Scientific Instrument Apr．2010 基于3种测度值的特征提取方法优化评价 王天杨，程卫东，李建勇 (北京交通大学机械与电子控制工程学院 北京 100044) 摘 要：模式识别结果的好坏往往取决于对识别对象本身特征的提取。对于一类对象通常已经存在若干成形的提取方法，要 想达到较好的识别分类效果就需要在特征提取方法的层面上获取更具代表性的特征表达。以字符识别所用的特征提取方法为 例，研究特征提取方法的分析评价。用5种常用的特征提取方法及其组合对标准字模进行特征提取，分别使用最小欧氏距离、 欧氏距离总和及欧氏距离方差3种测度在4个不同维度空间内对提取方法进行比较分析。最后，综合使用3种测度值去评价 特征提取方法。 关键词：模式识别；特征提取；欧氏距离；特征空间；综合测度 中图分类号：TP391．4 U491．116 文献标识码：A 国家标准学科分类代码：510．2040 Optimal evaluation of feature extraction method based on three types of measure values Wang Tianyang，Cheng Weidong，Li Jianyong (School ofMechanical，Electronic and Control engineering，Beijing Jiaotong University，Beijing 100044，China) Abstract：Whether the result of pattern recognition is good depends on the feature extraction of the objects being rec— ognized．There USUally exit some formed extraction．methods for the objects of a certain kind．So if better result of recognition and classification iS needed．more representative features should be extracted for the feature．extraction methods．In this paper，we take the feature extraction—method used in character recognition as an example to do the research on analysis and evaluation of feature extraction—methods．We use five common types of feature extraction· methods and their combinations to extract the features of the standard characters．Then we use three types of meas— ures to compare and analyze the extraction-methods in four different feature spaces respectively，which are minimum value of Euclidean distance，sum of Euc