基于正交分量辨别分析的人脸识别方法

ORTHONORMAL COMPONENT DISCRIMINANT ANALYSIS FOR FACE RECOGNITION

线性辨别分析(LDA)特征空间的坐标轴是非正交的,并且基于LDA的分类器的性能容易受训练集变化的影响。Okada提出了一种优化正交系统,但运算量大,且增加了特征空间坐标轴的数目,影响分类器速度。本文提出一种新的正交分量辨别分析(OCDA),没有增加坐标轴的数目,并且稳定性和识别率都优于LDA。

The LDA space is not orthonormal, and its performance is often affected by the variance of training set. Okada had put forward an optimal orthonormal system, but the number of coordinate axes of feature space was increased and its computation would spend a long time. A new orthonormal component discriminant analysis (OCDA) is proposed in this paper, in which the number of coordinate axes does not increase. The experiments on Yale and ORL databases show that OCDA is reliable and has a superior recognition rate to LDA.