中文摘要：

矩阵模式的Fisher线性判别准则（MatFLDA）作为近几年矩阵模式下的经典特征提取方法被广泛地加以研究和运用。然而MatFLDA方法作为全局判别准则一定程度上忽视了样本空间内在的局部结构和局部信息。在矩阵模式下，引入局部加权均值（LWM）并结合最大间距判别分析（MMC），提出一种具有局部学习能力的有监督的特征提取方法：基于矩阵模式的局部子域最大间距判别分析（Mat-LSMMC），提高了MatFLDA方法的局部学习能力，具有较强的特征提取能力。通过测试人造、真实数据集来表明所提方法的优势。

英文摘要：

MatFLDA as a classic feature extraction method in recent years is widely studied and used. However, MatFLDA as a global criterion is neglected to some extent sample space inner local structure and local information. Therefore, based on the ma trix pattern, by introducing the Local Weighted Mean （LWM） and combined with Maximum Margin Criterion （MMC）, we put forward a certain local learning ability of supervised feature extraction method： Local Sub - domains Maximum Margin Criterion Based Matrix Pattern, Mat - LSMMC）, the discrimination analysis with the strong ability of feature extraction can improve the lo cal learning ability of MatFLDA method. Finally, the test on artificial and real datasets shows the above mentioned advantages of the Mat- LSMMC method.