基于正则化协同表示的鲁棒PCA人脸识别

针对人脸图像复杂环境变化,提出了一种基于正则化协同表示的鲁棒PCA人脸识别算法。算法首先通过协同表示计算重构样本与测试样本之间的残差,根据残差选取与测试样本临近的训练样本组成新的字典;然后在该字典上通过鲁棒PCA依次进行低秩误差分解,并计算误差矩阵的平滑性、稀疏性;最后联合协同表示的残差以及低秩分解中的平滑性和稀疏性构建判别准则进行人脸图像的分类识别;在ORL、AR等人脸库的实验表明:基于正则化协同表示的鲁棒PCA人脸识别算法不仅能够在复杂环境变化下取得良好的识别性能,而且保持了协同表示的优势,大大减少运行时间。

Robust PCA Human Face Recognition Based on Regularized Collaborative Representation

According to complex environment changes of human face images, this paper proposes a robust PCA human face recognition algorithm based on regularized collaborative representation. This algorithm firstly uses collaborative representation to calculate the residuals between the reconstructed samples and test samples, selects the training samples nearing test samples based on the residuals to constitute a new dictionary, then uses robust PCA to conduct low rank and error decomposition in turn on this dictionary, calculates the smoothness and sparseness of the error matrix, and finally builds distinguishing criterion to conduct classified recognition of human face images by the residuals of collaborative representation and the smoothness and sparseness in low rank decomposition. The experiments on the databases such as ORL, AR and so on show that the robust PCA human face recognition algorithm based on regularized collaborative presentation can not only achieve good recognition performance under complex environment change but also maintain the advantages of collaborative presentation and largely reduce the computation time.