用于人脸识别的半监督优化局部保持投影

Semi-supervised Optimal Locality Preserving Projection for Face Recognition

未充分利用大量未标注样本的非监督信息是监督的、优化的局部保持投影(简称SOLPP)在人脸识别应用中的主要问题。为此提出一种用于人脸识别的半监督的优化的局部保持投影(SSOLPP)。该算法在SOLPP的基础上,通过加权平衡参数融合了未监督的主成分析(PCA)降维算法,使得投影后的数据保持了高维数据中的未标注样本的、全局的散布结构信息和监督的优化局部结构信息。在YaleB和AR人脸数据集上的实验验证了所提算法的有效性。

To ignore non-supervised information of unlabeled samples is the main problem of recently proposed Supervised Optimal Locality Preserving Projection （SOLPP） in applications of face recognition. Aiming to the prob- lem, a Semi-supervised Optimal Locality Preserving Projection （SSOLPP） for face recognition is proposed. On the base of SOLPP, the algorithm introduces Principal Component Analysis （PCA） with the weighted trade-off parame- ter way, making projected data preserve global scatter structure information and supervised optimal local structure information of high-dimensional data. Experimental results on Yale and YaleB demonstrate the effectiveness of our proposed algorithm.