组合2DFLDA监督的非负矩阵分解和独立分量分析的特征提取方法*

通过对投影非负矩阵分解(NMF)和二维Fisher线性判别的分析,针对NMF的特征提取存在无监督学习以及特征维数高的问题,提出了组合2DFLDA监督的非负矩阵分解和独立分量分析(SPGNMFICA)的特征提取方法。首先对样本进行投影梯度的非负矩阵分解,将得到的NMF子图像进行二维Fisher线性判别,主要反映类间差异信息构建子空间;对子空间的向量进行独立分量分析(ICA),得到独立分量特征空间;其次将样本在独立分量特征空间上进行投影;最后使用径向基网络对投影系数进行识别。通用人脸库ORL和YALE的识别实验证明,该算法是一种有效的特征提取和识别方法。

Feature extraction using 2DFLDA supervised projection gradient non-negative matrix factorization and ICA

With analysis of two-dimensional Fisher linear discriminant analysis and projection gradient non-negative matrix factorization(NMF), and in view of the existence of the NMF unsupervised learning and the high dimension problem, this paper proposed a novel feature extraction using 2DFLDA supervised projection gradient non-negative matrix factorization and ICA(SPGNMFICA). The method was first to employ projection gradient non-negative matrix factorization to samples. The subspace was constructed by vectors of mainly reflection difference between class, which was employed by ICA and the feature subspace was constructed. The sample was projected on the feature subspace. Lastly it classified the coefficient of projection by RBF. The experimental results on the ORL face database and YALE face database show that the proposed method is feasible and higher recognition performance.