摘要
针对人脸识别中的DCT系数选择问题和如何从全局和局部同时提取识别特征的问题,提出了一种基于鉴别能力分析和LDA-LPP的人脸识别算法。即先对人脸图像进行DCT变换,利用鉴别能力分析方法进行DCT系数的选择,融合LDA和LPP降维技术进行降维处理,不仅可以保持数据的全局性,同时也能够保持数据的局部性。在ORL人脸库和Yale人脸库上的实验表明,本文方法可以选择有效的DCT系数,明显提高了识别精度和鲁棒性。
The objectives of this paper are how to select the most effective DCT coefficients as recognition features and how to extract face features from locality and globality.A face recognition method is proposed based on Discrimination Power Analysis（DPA） and LDA-LPP algorithm.In this method,first,DCT transformation of face image is conducted.Then according to DPA,DCT coefficients are selected.Finally,integrating LDA,which can describe raw data from global situation,and LPP,which can maintain local adjacent relationship,a fusion recognition algorithm based on LDA-LPP is developed.The new algorithm can maintain both the local characteristics and global characteristics of the data.Besides,it rebuilds and lays out the raw data from the global situation.Experiments on ORL and Yale face databases show that the new algorithm can not only select the effective DCT coefficients but also improve the precision and robustness of the recognition process.
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2012年第6期1527-1531,共5页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金项目(61263031)
甘肃省自然科学基金项目(1010RJZA046)
甘肃省教育厅研究生导师基金项目(0914ZTB003)
甘肃省基本科研业务费项目(0914ZTB148)
关键词
计算机应用
鉴别能力分析
离散余弦变换
线性鉴别分析
局部保持投影
computer application
discrimination power analysis（DPA）
discrete cosine transform（DCT）
linear discriminate analysis（LDA）
local preserving projections（LPP）
