基于基空间转换的姿态人耳识别

Posed Ear Recognition Based on Basis Space Transformation

人耳识别是近年来新兴起的一种生物特征识别技术,然而姿态问题一直是其难点问题之一,针对此问题提出了一种基于基空间转换的新方法.首先,利用主元分析和核主元分析方法得到姿态人耳图像和正侧面人耳图像的基空间,通过计算两种基空间之间的线性转换关系求出姿态转换矩阵,然后将待测的姿态人耳图像特征集利用基空间姿态转换矩阵转变成正侧面人耳图像特征集,最后用支持向量机进行分类识别.实验结果表明,该方法与没有经过姿态转换的方法相比,识别率显著提高.

Ear recognition is a kind of new biometrics in recent years. Pose change, however, is one of the difficult problems to ear recognition. Therefore′ this paper proposes a new method based on basis space transformation to solve ear pose problem. Firstly, basis spaces including the posed ear images and the frontal ear images are obtained using PCA or KPCA. Secondly, the pose transformation matrix is gained in the light of linear transformation relation between two kinds of basis spaces. Thirdly, features of the posed ear images to be test are transformed into those of corresponding frontal ear images on the basis of the pose transformation matrix. Finally, the generated frontal ear features are identified by Support Vector Machine method. Experimental results show that the recognition rate with the pose transformation outperforms that without pose transfor- mation remarkably.