对称主分量分析及其在人脸识别中的应用

Symmetrical PCA and Its Application to Face Recognition

镜像对称性是人脸的一个直观显然的自然特性 ,有助于开发面向人脸图像的识别技术与算法 .本文将在人脸识别中应用这一自然特性 ,提出一种新算法———对称主分量分析 .该算法首先引入镜像变换 ,生成镜像样本 ;然后依据奇偶分解原理 ,生成镜象奇、偶对称样本 ,并分别进行K L展开 ,提取镜象奇 /偶对称KL特征分量 ;最后 ,根据奇 /偶对称KL特征分量在人脸中所占能量比例的不同以及对视角、旋转、光照等干扰的不同敏感程度进行特征选择 .理论分析与实验证明 ,该算法巧妙地利用镜像样本 ,增强人脸识别 :既扩大样本容量 ,显著提高识别率 ;又节省计算与存储开销 。

Facial symmetry is a useful natural characteristic of facial images. This paper will apply it to face recognition after introducing mirror images. By combining the K-L expansion with the even-odd decomposition principle, a new algorithm called Symmetrical Principal Component Analysis is proposed. In the algorithm, images are firstly decomposed into even symmetrical images and odd symmetrical ones. After that, even/odd symmetrical principal components are respectively extracted through K-L expansions, and then selected according to their energy ratios in faces and sensitivities to pattern variations. Both theoretical analysis and experimental results demonstrate that this algorithm has two outstanding advantages. Firstly, it remarkably raises the recognition rate. Secondly, it greatly saves the computational cost as well as the storage space.