摘要
为解决在人脸识别领域的特征提取问题,提出一种基于局部保持投影(LPP)的复合位置投影(MLPP)方法,通过选取不同的类内、类间度量矩阵和约束矩阵,将求解最优变换矩阵的问题转换成普通的特征值问题。在构造邻接图时,该算法将相同类各点作为邻接点,将类内结构保持到特征空间中,在保留局部结构稳定的同时,使整体结构趋于最大化,从而形成高效的聚簇。在AT&T和JAFFE标准人脸图像库上的实验结果表明,MLPP算法具有较高的识别率。
With the purpose of solving feature extraction problem in face recognition area, a new manifold learning algorithm is proposed, called Multiple Locality Preserving Projections （MLPP） based on Locality Preserving Projec- tions （LPP）. By selecting different measure matrix and constraints matrix those include intra-class matrix and inter-class matrix, the problem can be converted into the normal eigenvalue problem. When constructing the graph, this algorithm makes point with the same attributes as neighborhood points, which makes the intra-class construct save to feature space. As a result, the local construct remains stable, and at the same time the global construct tends to maximalism, so the cluster with high efficiency has been obtained. The results of the experiments on JAFFE and AT＆T face database indicate that MLPP improves recognition rate.
出处
《计算机工程与应用》
CSCD
2012年第32期208-211,共4页
Computer Engineering and Applications
基金
黑龙江省教育厅科学技术研究项目(No.11551087)
关键词
人脸识别
特征提取
局部保持投影
子空间
复合位置保持
face recognition
feature extraction
Locality Preserving Projections（LPP）
subspace
multiple locality
