摘要
边界Fisher判别分析算法因采用一维向量表示而无法很好保持图像的空间几何结构,且无法利用大量未标记样本信息。为此,提出一种基于张量的半监督判别分析算法。采用二维张量表示人脸空间中的样本图像,揭示流形的内在几何结构,利用有判别信息的标记样本和大量未标记样本,使数据在投影空间的类间分离度最大,同时保证高维空间中不相邻的点在低维空间中也不相邻。在PIE和FERET人脸库上的实验结果表明,该算法能够获得较高的识别率。
Marginal Fisher Analysis(MFA) algorithm is inadequate when it keeps the intrinsic geometric structure of images, and it only utilizes labeled data and wastes rich unlabeled data. This paper proposes a novel tensor subspaee learning algorithm: semi-upervised discriminant analysis algorithm based on tensor. The method adopts the two-dimensional tensor to show the image samples, so it can perfectly detect the intrinsic geometric structure of the data manifold. Moreover, it sufficiently utilizes the labeled data which contains discriminant information and rich unlabeled data, which can maximize the discriminant between classes of data in low dimension subspace and assure the distant points in the high dimension space is distant in the low-dimensional space. Experimental results on PIE and FERET face databases show that this algorithm can achieve higher recognition rate.
出处
《计算机工程》
CAS
CSCD
2012年第20期124-127,共4页
Computer Engineering
基金
甘肃省自然科学基金资助项目(0803RJZA109)
甘肃省科技攻关计划基金资助项目(2GS035-A052-011)
关键词
半监督判别
张量
流形学习
子空间
人脸识别
semi-supervised discriminant
tensor
manifold learning
subspace
face recognition