一种求解高维数据最佳鉴别向量的新算法

A NEW ALGORITHM FOR COMPUTING OPTIMAL DISCRIMINANT VECTORS IN CONDITION OF HIGH-DIMENSIONAL DATA

针对线性鉴别分析LDA(Linear Discriminant Analysis)方法在高维人脸图像识别领域的应用,提出一种计算最佳鉴别向量的新算法,无需对高维图像数据进行降维预处理,直接计算最佳鉴别向量。算法得到的鉴别向量相互正交,与已有的算法得到的鉴别向量相比,具有更好鉴别性能。在ORL和VALID人脸数据库上的实验结果证明了本算法的有效性。

For applying the Linear Discriminant Analysis （LDA） method to high-dimensional image data, especially for face recognition, we present a novel method to compute the optimal discriminant vectors without dimension reduction pre-processing on high dimensional image data and can directly compute the optimal discriminant vectors. Discriminant vectors calculated through our method are mutually orthonormal, and have higher discriminating performance comparing with those discriminant vectors derived from existing algorithms. Experimental results on ORL and VALID databases verify the effectiveness of our algorithm.