基于核主元分析法和支持向量机的人耳识别

对人耳识别中若干关键问题进行了研究. 介绍了两种人耳图像归一化处理的方法,即基于外耳轮廓长轴的线标记法和基于外耳轮廓起始点的点标记法,并对这两种方法进行了对比. 在分析现有人耳识别方法不足的基础上,提出利用核主元分析法提取人耳图像的代数特征,再利用支持向量机分类模型进行人耳识别. 在带有角度、光照变化的北京科技大学人耳图像库上得到的识别率为98.7%,表明了该识别方法的有效性以及利用人耳图像进行身份识别的可行性.

Ear recognition based on kernel principal component analysis and support vector machine

Some key issues in ear recognition were investigated. Two ear extraction and normalization methods, the mark line (long axis of the outer ear contour) based method and the mark points (the start and end points of the outer ear contour) based method, were proposed for recognizing ear images in the USTB ear database. Based on the analysis of the recent advances in ear recognition methods, the kernel principal component analysis (KPCA) was applied for ear feature extraction, and the support vector machine (SVM) model was applied for ear recognition. The ear recognition rate on USTB ear database with pose variation and lighting variation was 98.7%. The experimental result indicates the effectiveness of this method and proves the feasibility of ear recognition to be used in the field of personal authentication.