摘要： 传统的PCA 方法和LDA 方法在处理图像识别问题时，一般先将图像矩阵转化为图像向量，然后以该图像向量作为原始特征进行特征抽取。近来一些研究人员提出了利用图像矩阵直接构造散布矩阵，并在此基础上进行特征抽取的方法。该文在该思想的基础上，提出了IMSVD方法。该方法没有采用PCA 或LDA方法，而是利用奇异值分解方法进行特征抽取。对ORL 人脸图像的识别试验结果表明，IMSVD方法具有良好的特征抽取性能。

关键词: 图像识别；特征抽取；线性鉴别分析；主分量分析；奇异值分解；人脸识别

Abstract: Feature extraction is primary problem of image recognition. PCA and LDA are two classic methods applied widely in the field of image recognition. But they are both based on image vector in image recognition. Recently some researchers propose new methods which are based on image matrix. This paper proposes a new method called IMSVD. Based on the balanced scatter matrix, a discriminant criteria is formed. The optimal set of discriminant vectors can be acquired through singular value decomposition theorem. The result of face recognition experiment shows that it has powerful ability of feature extraction.

Key words: Image recognition; Feature extraction; Linear discriminant analysis; Principal component analysis; Singular value decomposition;Face recognition