现代计算机（普及版）
現代計算機（普及版）
현대계산궤（보급판）
MODERN COMPUTER
2014年
6期
43-47
,共5页
仲媛%杨健%涂庆华%李小舟
仲媛%楊健%塗慶華%李小舟
중원%양건%도경화%리소주
分类算法%KNN%均值%字符识别
分類算法%KNN%均值%字符識彆
분류산법%KNN%균치%자부식별
Classification Algorithm%KNN%Mean%Character Recognition
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。
基于傳統KNN的弱點，提齣一種新的改進方法：在訓練樣本中取與待測樣本最近鄰的K箇樣本；把這K箇樣本按類劃分取每類的平均樣本；計算每箇類的平均樣本與待測樣本的歐氏距離，併將樣本歸為距離最小的那一類。在南京理工大學NUST603HW手寫體漢字庫以及Concordia 大學的CENPARMI手寫體阿拉伯數字數據庫上的試驗結果錶明，新方法較傳統KNN的識彆率有明顯提高。
기우전통KNN적약점，제출일충신적개진방법：재훈련양본중취여대측양본최근린적K개양본；파저K개양본안류화분취매류적평균양본；계산매개류적평균양본여대측양본적구씨거리，병장양본귀위거리최소적나일류。재남경리공대학NUST603HW수사체한자고이급Concordia 대학적CENPARMI수사체아랍백수자수거고상적시험결과표명，신방법교전통KNN적식별솔유명현제고。
Based on the weakness of traditional KNN, proposes an improved method:takes K nearest neighbor samples of the tested sample in train-ing 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.
