KNN-均值算法  

KNN-Mean Algorithm

作　　者：仲媛[1] 杨健[2] 涂庆华[1] 李小舟[3] 

机构地区：[1]南京理工大学信息化建设与管理处,南京210094 [2]南京理工大学计算机科学与工程学院,南京210094 [3]南京航空航天大学发展计划处,南京210094

出　　处：《现代计算机（中旬刊）》2014年第6期43-47,共5页Modern Computer

基　　金：国家自然科学基金(No.60973098;No.60632050;No.90820306/F0307)

摘　　要：基于传统KNN的弱点,提出一种新的改进方法:在训练样本中取与待测样本最近邻的K个样本;把这K个样本按类划分取每类的平均样本;计算每个类的平均样本与待测样本的欧氏距离,并将样本归为距离最小的那一类。在南京理工大学NUST603HW手写体汉字库以及Concordia大学的CENPARMI手写体阿拉伯数字数据库上的试验结果表明,新方法较传统KNN的识别率有明显提高。Based on the weakness of traditional KNN, proposes an improved method： takes K nearest neighbor samples of the tested sample in training samples; divides these K samples by class and calculated the average of each class; calculates the Euclidean distance of the average sample of each class and the tested sample, and classified the tested sample into the class which is nearest. Carries out experiments at the Nanjing University of Science NUST603HW handwritten Chinese character library and Concordia University CENPARMI database of handwritten Arabic numerals library, the test results show that the recognition rate of the new method has improved significantly than the traditional KNN.

关 键 词：分类算法 KNN 均值 字符识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]