摘要： 为了克服PCA、ICA等传统方法在人脸图像特征抽取时存在速度慢、识别率低的缺点，该文提出了一种将非负矩分解思想应用于人脸特征提取的算法。利用小波变换对人脸图像进行分解，对其中包含主要信息的低频子带运用NMF构造特征子空间，在子空间内实现识别。实验结果表明，该方法实用、有效，减少了计算量，提高了系统的识别率，使识别率达到90％以上，有着广泛的研究价值和应用
前景。

关键词: 非负矩阵分解, 小波变换, 人脸识别, 子空间

Abstract: The traditional ways of character extraction of humane face image, such as PCA, ICA and so on, have low processing speed and low identification rate, this paper brings out a new method based on NMF in character extraction. An image is decomposed by using WT into different frequency subbands, then face recognition is implemented with the subspace. Experiment results indicate that this approach can largely reduce computing complexity and has higher recognition rate which is up to 90%. The results also show that the approach has obvious potential in practical usage.

Key words: Non-negative matrix factorization(NMF), Wavelet transform(WT), Face recognition, Subspace