摘要
针对传统KNN算法的手写数字识别运算量大和分类准确率低的问题,笔者提出一种基于分块LBP特征和训练集生成均值特征向量的改进KNN算法。该方法将样本图像分割为4个区域,分别提取各个区域的LBP直方图,然后将其组合作为该图像的特征向量。通过对比发现,基于分块LBP特征的传统KNN算法的分类准确率达到89%,但用时较长;运用改进后的KNN算法的数字分类准确率达到100%,同时大大减少了计算时间。
In view of the large amount of calculation and low classification accuracy of traditional KNN algorithm for handwritten digit recognition,the author proposes an improved KNN algorithm based on the block LBP feature and the training set to generate the mean feature vector.This method divides the sample image into 4 regions,extracts the LBP histogram of each region,and then combines them as the feature vector of the image.Through comparison,it is found that the classification accuracy of the traditional KNN algorithm based on the block LBP feature reaches 89%,but it takes a long time;the number classification accuracy of the improved KNN algorithm reaches 100%,and the calculation time is greatly reduced.
作者
赵腾浩
杨立娟
王宇
李晶
ZHAO Tenghao;YANG Lijuan;WANG Yu;LI Jing(School of Mechanical Engineering,Xi'an Jiaotong University,Xi'an Shaanxi 710049,China;School of Foreign Language,Southwest Minzu University,Chengdu Sichuan 610041,China)
出处
《信息与电脑》
2021年第5期195-200,共6页
Information & Computer
基金
西安交通大学2020年本科教学改革研究青年项目“制造产线虚拟仿真在实践教学中的应用研究”(项目编号:2002Q-07)
西安交通大学2019年本科实践教学改革研究专项项目“虚实结合的蒸压釜开门结构的应力应变测试实验”(项目编号:19SJZX18)。
关键词
手写体数字识别
改进KNN算法
分块局部二进制模式
handwritten number recognition
improved KNN algorithm
partitioned local binary pattern
