基于SVM的手写体数字快速识别方法研究
|
摘要
手写体数字识别是图像处理与模式识别中具有较高实用价值的研究热点之一。在保证较高识别精度的前提下,为提高手写体数字的识别速度,提出了一种基于SVM的快速手写体数字识别方法。该方法通过各类别在特征空间中的可分性强度确定SVM最优核参数,快速训练出SVM分类器对手写体数字进行分类识别。由于可分性强度的计算是一个简单的迭代过程,所需时间远小于传统参数优化方法中训练相应SVM分类器所需时间,故参数确定时间被大大缩减,训练速度得到相应提高,从而加快了手写体数字的识别过程,同时保证了较好的分类准确率。通过对MNIST手写体数字库的实验验证,结果表明该算法是可行有效的。
Handwritten digits recognition has high practical value in the field of image processing and pattern recognition. In order to improve the recognition speed,at the premise of high recognition accuracy,a fast handwritten digits recognition method based on SVM is proposed. The new method which uses the separability measure between classes in the feature space to choose the best kernel parameters, can train SVM classifiers fast to recognize the handwritten digits. Due to the computation of separability measure is a simple iterative process,the time required for computing is far less than the time required for training SVM classifiers in traditional parameter optimization methods. Thus,the time for kernel parameters selection will be reduced greatly. Accordingly,the training speed will be increased,and so that the process of recognizing handwritten digits will also be speeded up,while ensuring better classification accuracy. The experiment results of testing MNIST show that the improved algorithm is feasible and effective.
Handwritten digits recognition has high practical value in the field of image processing and pattern recognition. In order to improve the recognition speed,at the premise of high recognition accuracy,a fast handwritten digits recognition method based on SVM is proposed. The new method which uses the separability measure between classes in the feature space to choose the best kernel parameters, can train SVM classifiers fast to recognize the handwritten digits. Due to the computation of separability measure is a simple iterative process,the time required for computing is far less than the time required for training SVM classifiers in traditional parameter optimization methods. Thus,the time for kernel parameters selection will be reduced greatly. Accordingly,the training speed will be increased,and so that the process of recognizing handwritten digits will also be speeded up,while ensuring better classification accuracy. The experiment results of testing MNIST show that the improved algorithm is feasible and effective.
引文
[1]边肇祺,张学工.模式识别[M].北京:清华大学出版社,2000.
[2]Vapnik V N.统计学习理论[M].许建华,张学工,译.北京:电子工业出版社,2004.
[3]Astorino A,Gorgone E,Gaudioso M,et al.Data preprocessing in semi-supervised SVM classification[J].Optimization,2011,60(1-2):143-151.
[4]Hu Guyu,Gong Yong,Chen Yande,et al.Semi-supervised radio transmitter classification based on elastic sparsity regularized SVM[J].Journal of electronics(China),2012,29(6):501-508.
[5]李文趋.SVM在手写数字识别中的应用[J].泉州师范学院学报(自然科学),2010,28(4):18-21.
[6]石会芳,胡小兵,刘瑞杰,等.基于启发式GA-SVM的手写数字字符识别的研究[J].计算机技术与发展,2012,22(10):5-9.
[7]蒋华,戚玉顺.基于球结构SVM的多标签分类[J].计算机工程,2013,39(1):294-297.
[8]刘端阳,邱卫杰.基于SVM期望间隔的多标签分类的主动学习[J].计算机科学,2011,38(4):230-232.
[9]Hsu C W,Lin C J.Acomparison of methods for multi-class support vector machines[J].IEEE transactions on neural networks,2002(13):415-425.
[10]奉国和.SVM分类核函数及参数选择比较[J].计算机工程与应用,2011,47(3):123-124.
[11]王佳,徐蔚鸿.基于动量粒子群的混合核SVM参数优化方法[J].计算机应用,2011,31(2):501-503.
[12]陈圣兵,王晓峰.基于样本差异度的SVM训练样本缩减算法[J].计算机工程与应用,2012,48(7):20-22.
[13]王涛,程良伦.基于快速SVM的大规模网络流量分类方法[J].计算机应用研究,2012,29(6):2301-2305.
[14]Chang C C,Lin C J.LIBSVM:A library for support vector machines[EB/OL].2001[2013-03-04].http://www.csie.ntu.edu.tw/~cjlin/papers/libsvm.pdf.
[2]Vapnik V N.统计学习理论[M].许建华,张学工,译.北京:电子工业出版社,2004.
[3]Astorino A,Gorgone E,Gaudioso M,et al.Data preprocessing in semi-supervised SVM classification[J].Optimization,2011,60(1-2):143-151.
[4]Hu Guyu,Gong Yong,Chen Yande,et al.Semi-supervised radio transmitter classification based on elastic sparsity regularized SVM[J].Journal of electronics(China),2012,29(6):501-508.
[5]李文趋.SVM在手写数字识别中的应用[J].泉州师范学院学报(自然科学),2010,28(4):18-21.
[6]石会芳,胡小兵,刘瑞杰,等.基于启发式GA-SVM的手写数字字符识别的研究[J].计算机技术与发展,2012,22(10):5-9.
[7]蒋华,戚玉顺.基于球结构SVM的多标签分类[J].计算机工程,2013,39(1):294-297.
[8]刘端阳,邱卫杰.基于SVM期望间隔的多标签分类的主动学习[J].计算机科学,2011,38(4):230-232.
[9]Hsu C W,Lin C J.Acomparison of methods for multi-class support vector machines[J].IEEE transactions on neural networks,2002(13):415-425.
[10]奉国和.SVM分类核函数及参数选择比较[J].计算机工程与应用,2011,47(3):123-124.
[11]王佳,徐蔚鸿.基于动量粒子群的混合核SVM参数优化方法[J].计算机应用,2011,31(2):501-503.
[12]陈圣兵,王晓峰.基于样本差异度的SVM训练样本缩减算法[J].计算机工程与应用,2012,48(7):20-22.
[13]王涛,程良伦.基于快速SVM的大规模网络流量分类方法[J].计算机应用研究,2012,29(6):2301-2305.
[14]Chang C C,Lin C J.LIBSVM:A library for support vector machines[EB/OL].2001[2013-03-04].http://www.csie.ntu.edu.tw/~cjlin/papers/libsvm.pdf.