摘要
使用RBF核的SVM(支持向量机)被广泛应用于模式识别中。此类SVM的模型选择取决于两个参数,其一是惩罚因子C,其二是核参数σ2。该文使用了网格搜索和双线性搜索两种方法进行参数选择,并将两者的优点综合,应用于脱机手写体英文字符识别。实验在NIST数据集上进行了验证,对搜索效率和推广识别率进行了比较。实验结果还表明使用最优参数的SVM在识别率上比使用ANN(人工神经元网络)的分类器有较大提高。
SVM(Support Vector Machine)with RBF kernel is widely used in pattern recognition.Model selection in this class of SVMs involves two parameters:the penalty parameter C and the kernel parameter.This paper uses grid search and two-line search to select the two parameters,and combines the advantage of the two methods to the application on Handwritten English Character Recognition.Experiments are performed on the NIST database to acquire the comparison of the searching efficiency and generalized recognition rate.It's also shown that SVM with the best parameters are much better on generalized regcognition rate against the ANN(Artificial Neural Network)classifier.
出处
《计算机工程与应用》
CSCD
北大核心
2003年第24期72-73,共2页
Computer Engineering and Applications
基金
国家重点基础研究发展规划项目资助(编号:G199803050703)
国家自然科学基金资助(编号:60272019)
关键词
SVM
RBF核
模型选择
ANN
字符识别
Support Vector Machine(SVM),RBF kernel,Model Selection,ANN,Character Recognition