摘要
研究、分析了人脸识别中提取原始数据特征的已有方法,在此基础上给出了一种应用监督式正交迹比判别投影(SOTRDP)的新型特征提取方法,即SOTRDP方法。不同于现有的非监督判别投影(UDP)方法,SOTRDP方法能够同时利用局部信息和类别信息建立相似性矩阵。在利用改进局部切空间对齐(ILTSA)非线性降维的基础上,利用聚类中心或最靠近它的样本作为输入,拓展SOTRDP用于图像集人脸识别。在PIE和Honda/UCSD人脸数据库上的实验结果验证了所提方法的有效性。
The existing feature extraction methods for face recognition were studied and analyzed, and based on this, a novel feature extraction method using supervised orthogonal trace ratio discriminant projection （SOTRDP）, called the SOTRDP method for short, was proposed. Unlike the unsupervised discriminant projection （UDP） method, this new method can simultaneously use both local information and class information to model the similarity of the data. Based on the improved local tangent space alignment （ILTSA） nonlinear dimensionality reduction, the SOTRDP was extended to image set face recognition application by using the clustering centers as input. The results of the experiments on the PIE and Honda/UCSD image set face databases demonstrate the effectiveness of the proposed face recognition method.
出处
《高技术通讯》
CAS
CSCD
北大核心
2014年第7期684-689,共6页
Chinese High Technology Letters
基金
国家自然科学基金(61004088)
上海市基础研究重点(09JC1408000)资助项目
