关键词: 线性判别分析（LDA）, 离散余弦变换（DCT）, 直接LDA法（Direct LDA）, 小样本问题（SSS）

Abstract: D-LDA is a simple and efficient linear projection technique for feature extraction.But it may encounter two probloms in application：（1）it may indirectly misses some useful information in null space of Sw while deleting null space of Sb;（2）its optimization criteria is not directly related to the classification accuracy.Discrete Cosine Transform（DCT） can compress the information of original signal efficiently.We propose an approach based on DCT and improved D-LDA.We reduce the dimension first and then extract features by improved D-LDA on the low dimension space to overcome the shortcomings of D-LDA furthest.The experimental result will show that this method has a better performance.

Key words: Linear Discriminant Analysis（LDA）, Discrete Cosine Transform（DCT）, D-LDA, Small Sample Size problem（SSS）