摘要
为快速有效地在掌纹识别中学习多种因素的高阶统计独立成分,利用多线性独立成分分析方法对掌纹张量进行降维,得到低维的模式矩阵,将掌纹图像向模式矩阵上投影以提取核心张量,通过计算核心张量间的余弦距离实现掌纹匹配。基于PolyU掌纹图像库的实验结果表明,与主成分分析(PCA)、二维PCA、独立成分分析和多线性PCA相比,该方法的识别率最高,且满足系统实时性要求。
In palmprint recognition, in order to learn the higher order dependencies associated with the different factors quickly and effectively, this paper uses Multilinear Independent Component Analysis(MICA) to the original tensor objects to obtain the low-dimension mode matrix. The palmprint images are projected onto the mode matrix for extracting the core tensors. Palmprint matching is implemented by calculating the cosine distance between two core tensors. Experiment based on PolyU plmprint database shows that compared with Principal Component. Analysis(PCA), 2DPCA, Independent Component Analysis(ICA) and Multilinear Principal Component Analysis(MPCA), recognition rate of the new algorithm is the highest, and it meets real-time requirements of the system.
出处
《计算机工程》
CAS
CSCD
北大核心
2011年第12期13-15,18,共4页
Computer Engineering
基金
国家自然科学基金资助项目(60774070)
辽宁省科研基金资助项目(L2010436
20091061)
关键词
掌纹识别
主成分分析
二维主成分分析
多线性主成分分析
独立成分分析
多线性独立成分分析
palmprint recognition
Principal Component Analysis(PCA)
2D Principal Component Analysis(2DPCA)
Multilinear PrincipalComponent Analysis(MPCA)
Independent Component Analysis(ICA)
Multilinear Independent Component Analysis(MICA)