【摘要】 手写数字识别在很多领域都有着广泛的应用前景,论文研究了Gabor滤波器的特性及其特征提取的方法,PCM特征向量选择降低数据维数,SVM支持向量机的原理以及影响其性能的参数,并且克服了传统的维数灾难与过学习现象,用MNIST数据库做了仿真实验,交叉验证取得合适的SVM参数.实验表明,该方法取得了比较理想的结果,验证了SVM的多类分类方法,并与传统的一些分类方法做了比较分析.更多还原

【Abstract】 Handwritten digit recognition has wide application foreground in many domains.This paper studied the character of Gabor filter,the methods of feature extractor,principal component analysis,the theory of SVM and its parameters.The method proposed overcame dimension disaster and overfitting.The experiment with MNIST database validated there were ideal results with our method,and also the multi-class SVM was efficient.At last,we analyzed and compared the methods with traditional multi-class classification.更多还原