图像识别中的2维线性鉴别分析(2DLDA)实际上是将图像的各个列(或行)视为样本向量，但这些样本向量不能满足统计学中的独立同分布要求。为克服2DLDA的不足，提出了基于图像抽样重组的2DLDA (SR2DLDA)，它对图像进行下抽样，并将抽样所得的不同小图像重组成矩阵，然后对这些矩阵实施2DLDA。由于抽样重组的矩阵改善了各个列向量的独立性与分布同一性，因而SR2DLDA的识别性能有可能优于2DLDA，也优于LDA。在ORL人脸库、UMIST人脸库和FERET人脸库上的实验验证了SR2DLDA的有效性。

The columns or rows of an image are practically viewed as sample vectors in two dimension linear discriminent analysis (2DLDA). However, those sample vectors can not fulfill the independent identically distributed requirement in statistics. This paper proposes a method, called Sampling and Regroupment 2DLDA (SR2DLDA), which can improve 2DLDA and LDA. SR2DLDA down-samples the sample images, regroups the small down-sampling images to matrices, and then performs 2DLDA on them. These matrices may make progress on the independent identically distributed requirement. The experiments on ORL database, UMIST database and FERET database verify the efficiency of the SR2DLDA.