基于局部保持投影和核直接判别分析的掌纹识别  

Palmprint recognition based on locality preserving projection and kernel direct discriminant analysis

作　　者：郭金玉[1] 李元[1] 孔晓光[1] 曾静[1] 

机构地区：[1]沈阳化工大学信息工程学院,辽宁沈阳110142

出　　处：《光电子．激光》2011年第1期127-130,共4页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金资助项目(60774070);辽宁省教育厅科研资助项目(L2010436);辽宁省博士启动基金资助项目(20091061)

摘　　要：为了提高识别性能,提出运用局部保持投影(LPP)和核直接判别分析(KDDA)相结合的方法进行掌纹识别。在小样本图像识别中,为了解决特征方程矩阵的奇异性,首先运用图像下抽样降低掌纹空间的维数,然后应用LPP提取掌纹局部结构特征作为KDDA的输入提取分类特征,计算特征向量间的余弦距离进行掌纹匹配。运用PolyU掌纹图像库,对本文算法进行测试。实验结果表明,与主元分析(PCA)、独立元分析(ICA)、PCA+LPP、核局部保持投影(KLPP)、核判别分析(KDA)和抽样(sample)+LPP相比,本文算法的识别率(RR)最高为99.71%,特征提取和匹配总时间为0.131 s,满足实时系统的要求。A palmprint recognition method based on locality preserving projection（LPP） and kernel direct discriminant analysis（KDDA） was proposed in order to improve the recognition performance.In small-size-sample cases（such as image recognition）,in order to resolve the singularity of the eigenvalue equation,the image down-sample was first applied to reduce the palmprint space dimension,the LPP was used to extract the local structure features as the input classification extraction features of KDDA,and the cosine distance between two feature vectors was calculated to match palmprints.The new algorithm was tested with PolyU plmprint database.The experimental results show that compared with principal component analysis（PCA）,independent component analysis（ICA）,PCA＋LPP,kernel locality preserving projection（KLPP）,kernel direct discriminant analysis（KDDA） and sample＋LPP,the recognition rate（RR） of the proposed algorithm is the highest up to 99.71%,and the total time for feature extraction and matching is only 0.131 s,so it meets the real-time system requirement.

关 键 词：图像处理 掌纹识别 下抽样 局部保持投影(LPP) 核直接判别分析(KDDA) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]