基于局部二进制和支持向量机的手写体数字识别  

Handwritten digit recognition based on local binary pattern and support vector machine

作　　者：郭伟林 邓洪敏[1] 石雨鑫 GUO Weilin;DENG Hongmin;SHI Yuxin(College of Electronics and Information Engineering,Sichuan University,Chengdu Sichuan 610065 China)

机构地区：[1]四川大学电子信息学院,成都610065

出　　处：《计算机应用》2018年第A02期282-285,289,共5页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61174025)

摘　　要：针对手写体数字识别中的特征提取问题,在圆形邻域局部二进制模式(LBP)的基础上,将数字的图片分割为四个子区域,分别提取各区域的局部二进制模式(LBP)直方图,然后用LBP直方图训练支持向量机(SVM)模型,再用测试样本对该模型的分类效果进行测试。最后,在实验中还引入当前主流的主成分分析法(PCA)对数据集降维,并以此对比分四区域提取LBP直方图的识别效果。通过对比发现,对手写体数字图片分四区域计算LBP直方图并将其用作识别特征可使数字识别准确率达到95. 31%,这一结果接近于使用数据集前90%特征贡献率的主成分分量的识别效果(后者的识别率为96. 6%),比直接对整张图片提取的LBP直方图分类识别率高19. 51个百分点。To extract the features of digit image in handwritten digital recognition,the digital images were divided into four subregions,and then the Local Binary Pattern(LBP)histogram of each region was extracted respectively,which was used to train Support Vector Machine(SVM)model,and then the classification performance of the model was evaluated by using the test samples.In addition,Principal Component Analysis(PCA)algorithm was used to compare the experimental results.The experimental results show that dividing a digit image into four subregions and calculating its LBP histogram can achieve the accuracy of 95.31%in digit recognition,which is close to the recognition result of 96.6%that preserves principal components of the first 90%feature contribution rate of the PCA dimensionality reduction,and also 19.51 percentage points higher than that of LBP histogram extracted from the whole image.

关 键 词：手写体数字识别 局部二进制模式 局部二进制模式直方图 支持向量机 主成分分析法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]