基于核稀疏表示的多模身份识别算法

Multi-modal identification algorithm based on kernel sparse representation

针对训练样本与测试样本非线性可分问题,借助核算法,将样本特征向量映射到易实现线性可分的核空间,进而在高维核空间内运用核稀疏表示对所提取的特征进行分类表达。该算法受益于将核稀疏表示理论同多模生物识别技术相结合,使其对生物特征图像有较强的鲁棒性。实验证明基于核稀疏表示的多模身份识别算法在遮挡、含噪声的情况下具有较好的识别准确率,相较于其他同类算法在性能上有一定程度的提高。

Aiming at the problem of non-linear separability between training sample and test sample,this paper uses the kernel arithmetic to map the sample eigenvector to the kernel space which is easy to realize linear separability,and then the kernel sparse representation is used in the high-dimensional kernel space to classify the extracted features.This algorithm benefits from the combination of kernel sparse representation theory with multi-modal biometrics,which results in robustness to biometric images.Experiments show that the multi-modal identification algorithm based on kernel sparse representation has better recognition accuracy under occlusion and noises,which improves the performance to some extent compared with other similar algorithms.