基于结构化低秩表示和低秩投影的人脸识别算法

在实际的人脸识别中，给定的训练图像往往存在遮挡和噪声，导致稀疏表示分类（SRC）算法的性能下降。针对上述问题，提出一种基于结构化低秩表示（SLR）和低秩投影的人脸识别方法--SLR_LRP。首先通过SLR对原始训练样本进行低秩分解得到干净的训练样本，根据原始训练样本和恢复得到的干净训练样本得到一个低秩投影矩阵；然后将测试样本投影到该低秩投影矩阵；最后使用SRC对恢复后的测试样本进行分类。在AR人脸库和Extended Yale B人脸库上的实验结果表明，SLR_LRP可以有效处理样本中存在的遮挡和像素破坏。

Face recognition based on structured low rank representation and low rank projection

Occlusion and corruption in the training images result in degraded performance of the sparse representation classification (SRC) algorithm in practical applications of face recognition. Aiming at the aforementioned problem, we propose a new face recognition method based on structured low rank representation (SLR) and low rank projection (LRP), called SLR_LRP. Firstly, the original training samples are decomposed via SLR to obtain clean training samples. And a LRP matrix is learned based on the original training samples and the recovered clean samples. Secondly, test samples are projected onto the LRP matrix. Finally, SRC is exploited to classify the corrected test samples. Experiments on the AR and the Extended Yale B face databases demonstrate that the SLR_LRP can effectively deal with the occlusion and pixel corruption in samples.