基于小波和非负稀疏矩阵分解的人脸识别方法

Wavelet-based Non-negative Matrix Factorization with Sparseness Constraints for Face Recognition

提出了利用小波变换(WT)、非负稀疏矩阵分解(NMFs)和Fisher线性判别(FLD)来进行人脸识别。用小波变换分解人脸图像,选择最低分辨率的子段,既能捕获到人脸的实质特征,又有效地降低了计算复杂性;非负稀疏矩阵分解能显示地控制分解稀疏度和发现人脸图像的局部化表征;Fisher线性判别能在低维子空间中形成良好的分类。实验结果表明,这种方法对光照变化、人脸表情和部分遮挡不敏感,具有良好的健壮性和较高的识别效率。

This paper combines Wavelet Transformation（WT）, Non-negative Matrix Factorization with sparseness constraints （NMFs）, and Fisher＇s Linear Discriminant （FLD） to extract features for face recognition. Wavelet transformation is used to decompose face images and for choosing the lowest resolution sub-band coefficients so that the substantial facial features can be captured and the computational complexity can be reduced. NMFs can control sparseness explicitly and find parts-based representations for face images. FLD plays the role of forming well-separated classes in a low-dimensional subspace. Extensive experiments are carried out to illustrate the proposed combine face recognition method by using the ORL face database. The experimental results show that the method has robust high-performance against varying illumination, facial expression and part occlusion.