一种基于核函数特征提取改进方法的应用  

Feature extraction based on kernel function and application of improved methods

作　　者：李德启[1] 刘传领[2] 

机构地区：[1]商丘职业技术学院,河南商丘476000 [2]南京理工大学计算机学院,南京210094

出　　处：《计算机应用研究》2011年第8期3185-3187,3191,共4页Application Research of Computers

摘　　要：针对标准的核典型相关分析(KCCA)方法在对训练样本增大的情况下相应计算机复杂度剧增、内存占用量大的缺陷,在对标准的KCCA特征提取方法分析推导的基础上,提出了一种改进的核函数特征提取方法。该方法首先根据特征值的大小对训练样本重要程度进行判断,进而完成对应特征向量的提取;然后通过与SVDD分类器的结合,在对图像识别率影响不大的情况下,提升了对图像特征提取的效率,节省了系统的存储量;最后通过在Yale标准人脸库上进行仿真对比实验,验证了该方法的可行性,从而为提高图像模式识别效率提供了一种有效的途径。For the standard KCCA method in the case of training samples increases the complexity of the corresponding surge in the computer memory occupied by a large quantity of defects,based on the standard derivation of KCCA feature extraction methods,this paper proposed an improved feature extraction method of nuclear function.In this method,the value based on characteristics of the training sample size to judge the degree of importance,and then completed the corresponding eigenvectors,and combined with the SVDD classifier,the image recognition rate in the case of little effect to enhance the features of the image extraction efficiency,saving the system＇s storage capacity,and finally by the Yale face database standard simulation comparison experiments were carried out to verify the feasibility of the method.Thus improves the efficiency of image pattern recognition provides an effective way.

关 键 词：核典型相关分析 特征提取 计算复杂度 内存占用量 识别率 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]