联合松弛块对角表示矩阵回归的遮挡人脸识别

基于核范数矩阵回归的方法能够有效解决人脸识别中连续遮挡的问题，然而该类方法仅关注误差图像的低秩结构信息，忽略了样本图像表示的相关性。为了有效解决自然场景下的遮挡人脸识别问题，考虑到这一特点，提出一种联合松弛块对角表示的矩阵回归模型（RBDMR）学习图像的松弛块对角表示，并通过动态优化表示矩阵的块对角分量加强类内表示的相关性和类间表示的差异性。此外，通过联合优化训练样本和测试样本的表示持续提高类内表示的一致性。通过在三个不同的数据集进行验证，实验结果表明，该方法优于其他对比算法，在真实遮挡和光照变化的情况下有较好的性能。

Occlusion face recognition with relaxed block diagonal representation matrix regression

Nuclear norm based matrix regression method can effectively solve the continuous occlusion problem in face recognition. However, these methods only focus on the low-rank structural of the error images, which ignores the correlation of sample images representation. In order to effectively solve the occluded face recognition problem in natural scenes, this paper proposed the matrix regression model with joint relaxed block-diagonal representation(RBDMR) to learn the relaxed block-diagonal representation of images, then strengthened the correlation of the intra-class representation and the differences of the inter-class representation by dynamically optimizing the block-diagonal component of the representation matrix. Furthermore the coherence of the intra-class representation was continuously improved by jointly optimizing the representation of training samples and test samples. Through verification on three different databases, the experimental results show that the proposed method outperforms other comparison algorithms and has better performance in the real occlusion and illumination changes.