计算机科学与探索 ›› 2011, Vol. 5 ›› Issue (11): 1021-1026.

• 学术研究 • 上一篇    下一篇

Finsler度量在KNN算法中的应用研究

陈 明, 何书萍, 李凡长   

  1. 苏州大学 计算机科学与技术学院, 江苏 苏州 215021
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2011-11-01 发布日期:2011-11-01

Research on Finsler Metric in KNN Algorithm

CHEN Ming, HE Shuping, LI Fanzhang   

  1. College of Computer Science and Technology, Soochow University, Suzhou, Jiangsu 215021, China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-11-01 Published:2011-11-01

摘要: 为了克服传统K近邻(K nearest neighbor, KNN)算法在距离定义上的不足, 提出了一种基于Finsler度量的KNN算法(Finsler metric KNN, FMKNN)。该算法将样本点间的距离定义为Finsler度量, 保留了样本属性对样本间距离度量的影响, 使得样本点间的距离度量更具一般性。在手写体数据集上的实验表明, FMKNN算法的分类准确率高于传统KNN算法。

关键词: K近邻(KNN), Finsler度量, 手写体识别

Abstract: In order to overcome the shortcomings of traditional K nearest neighbor (KNN) algorithms in distance definition, this paper proposes a new KNN algorithm based on Finsler metric, FMKNN. The algorithm defines the distance between sample points as the Finsler metric and preserves the distance between sample properties, making the distance between sample points more general. The experiment on handwritten data sets shows that, the classification accuracy of FMKNN algorithm is higher than traditional KNN algorithms.

Key words: K nearest neighbor (KNN), Finsler metric, handwriting recognition