基于低秩投影与稀疏表示的人脸识别算法

当训练和测试图像同时受到污损时,人脸识别的性能会急剧下降。为了解决这一问题,提出了一种新的人脸识别算法。首先利用鲁棒主成分分析(robust principal component analysis,RPCA)方法得到训练样本的低秩部分;然后基于原始训练样本及其低秩部分得到低秩投影矩阵,该矩阵可以对存在污损的测试图像进行恢复;最后使用稀疏表示分类(sparse representation based classification,SRC)算法对恢复后的测试图像进行分类。在两个公开数据库上进行实验,实验结果证明了本文算法的有效性,同时识别性能优于SRC及线性回归分类(linear regression classification,LRC)方法,能在一定程度上处理样本数据受到污损的情况。

Face Recognition based on Low-Rank Projection and Sparse Representation

When providing corrupted training and test samples, performance of face recognition will degrade dramatically. To mitigate this problem, a new method for face recognition is proposed. Firstly, the training data is decomposed via RPCA to obtain its low rank part, then a low-rank projection matrix is learned based on the original training data and its low rank part. This projection matrix is capable of correcting corrupted test images. Finally the corrected test samples are classified based on SRC. Experimental results on two publicly available databases document the effectiveness of the proposed method, and it achieves better performance than SRC based approaches and LRC, meanwhile it can handle the case that samples are contaminated to some extent.