基于特征融合的人脸人耳多生物身份鉴别

针对人头旋转时单一人脸识别率较差的问题，基于人脸与人耳位置上的关联性，提出人脸人耳多生物特征级融合的身份鉴别方法以克服姿态带来的影响．首先采用传统独立成分分析（ICA）方法及其变形分别提取出图像的局部和全局特征，然后将这2种互补的特征进行多模态加权串联融合，并采用基于非线性核函数的主元分析法（KPCA）降维．在USTB图像库上的实验表明，2种独立成分特征具有很好的互补性，多生物识别大大优于单一生物识别，且提出的核非线性降维方法进一步改善了识别性能．

Face and Ear Feature Fusion for Human Recognition

The performance of face recognition is poor when there is a large pose variation. In view of the correlation of face and ear, a method of human identification by the feature fusion of face and ear was proposed. Firstly ,the local and global features were extracted respectively by conventional independent component analysis （ICA）and its improved method,then the features from face and ear were fused by series weighed strategy and the dimension of new feature was descended by kernel principal component analysis （KPCA）. Experiments on USTB database show that the two types of ICA features are complementary,multi-biometrical system performs much better than single modality,and the non-linear descending dimension method improves the recognition rate in some degree.