基于物体特征轮廓的单类判别方法  

One-Class Classification Based Eigen Contours

作　　者：许强[1] 马登武[1] 

机构地区：[1]海军航空工程学院兵器科学与技术系,烟台264001

出　　处：《计算机辅助设计与图形学学报》2014年第6期932-938,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：航空科学基金(20095184004)

摘　　要：针对当前物体轮廓分类方法的不足,提出一种基于物体特征轮廓的单类判别方法.首先修正了一种利用傅里叶描述子对轮廓归一化的方法;然后采用主成分分析对训练轮廓集的符号中心距离点列进行特征向量提取,生成类似"特征脸"的特征轮廓;最后根据轮廓在特征轮廓上投影所产生的截断误差不同设定阈值,进行轮廓类别判定.实验结果表明,与单类支持向量机方法相比,该方法的判别准确率提高了20%.Aiming at the shortcomings of the current object contour classification methods,we presented a novel method of one-class classification based on eigen contour.First a common normalization method of object contour using Fourier descriptor is corrected; Then eigen contour is created,which is similar with ＂eigen face＂ by applying PCA to the point range of signed centroid distance; At last,the contour according to the threshold value is classified,which is based on the principle that different kind of contour has different projecting value on the eigen contour.Experimental results verify the effectiveness of this method with classification accuracy 20％ higher than that of one-class SVM.

关 键 词：特征轮廓 单类判别 傅里叶描述子 符号中心距离 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]