摘要
为了进一步提高零空间算法的运行效率,提出了一种新的快速的零空间算法(FINBSA).FINBSA不需要进行特征值分解或奇异值分解,而只需一次正交三角(QR)分解就可以求得最佳投影矩阵,使得FINBSA的算法复杂度比现有的零空间算法要低.在PIE人脸库上的实验结果表明,FINBSA的识别率与现有的零空间算法相同,但是远比现有的零空间算法要高效,尤其是在训练样本数较多时,FINBSA的运行时间比现有零空间算法节省了100%以上.
To improve the efficiency of null space based algorithm,a new fast implementation of null space based algorithm（FINBSA） is presented.FINBSA is carried out without any eigen-decomposition computing and singular value decomposition,by only one step of orthogonal-triangular（QR） decomposition to obtain the optimal projection matrix.Thus the computational complexity of this algorithm gets much lower than that of the other null space based ones.The experiments on PIE face database demonstrate that the recognition accuracy of FINBSA is equivalent to the other null space based algorithms,but FINBSA is more efficient especially for large size of training samples,and the running period of FINBSA is over 100% lower than that of the other existing null space based algorithms.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2012年第2期59-63,共5页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(60873151)
关键词
特征提取
线性鉴别分析
零空间线性鉴别分析
正交三角分解
人脸识别
feature extraction
linear discrimination analysis
null space based linear discrimination analysis
orthogonal-triangular decomposition
face recognition
