基于支持向量机的英文字符识别研究-《计算机技术与发展》

文章信息/Info

Title:: Research on English Character Recognition Based on Support Vector Machine

Author(s):: ZHENG Yu-chen; School of Mathematics and Software Sciences,Sichuan Normal University,Chengdu 610068,China

Keywords:: handwritten English character recognition; data mining; radial basis function (RBF); MCSVM; statistical machine learning; penalty parameter C

摘要:: 图像识别是"大数据"时代的热门研究领域之一,而英文字符识别是图像识别领域重要的研究方向.对于手写数据的辨认在移动智能、刑侦、医学、考古学等诸多领域有广泛的应用,同时,国内在该领域的建模探索相对匮乏.文中使用机器学习领域的经典手写字符数据集,基于统计机器学习理论,建立英文字符识别的支持向量机(SVM)模型.鉴于国内外对于参数选择至今没有公认的方法,依据支持向量的个数、训练误差、测试误差作为评价指标,对惩罚参数C的选取进行探索并给出了在字符识别领域的推荐值.实证结果表明,对"变体"英文字母的识别准确率很高,且非常稳健,没有"过拟合"现象,说明支持向量机适用于处理字符识别问题.本质上,相比经典的二分类问题,文中是多分类支持向量机(multi-class classification support vector machine,MCSVM)应用的研究与探索.

Abstract:: Pattern recognition is one of the hottest research fields in the era of “Big Data”,and English character recognition is considereda significant research orientation of pattern recognition. Recognizing handwritten data is widely used in many fields,such as mobile intel-ligence,criminal investigation,medicine,and archaeology. However,there is rare domestic modeling research in this field. Based on thestatistical machine learning theory,we make use of a classic handwritten data set in the field of machine learning to build a support vectormachine (SVM) model of English character recognition. It is well-known that there isn’t any widely accepted method to select parame-ters for SVM even in foreign articles. According to the fact,research on how to select penalty parameter C is implemented based on theindex like,the number of support vector,training error and test error. More than that,a recommended penalty parameter C of letter recog-nition is also proposed. The experiment indicates that this model has high accuracy and robustness without overfitting to recognize varia-tion English character. So SVM is a favorable choice to handle with character recognition. Essentially,this article aims at applied re-search and exploration of the multi-class classification support vector machine (MCSVM) compared with classical binary SVM.