基于迭代加权低秩分解的遮挡人脸识别算法

An Occlusion Face Recognition Algorithm Based on Iteratively Reweighted Robust Principal Component

针对传统低秩矩阵分解算法得出的稀疏矩阵中既包含遮挡因素和噪声因素的问题,提出基于迭代加权低秩分解的遮挡人脸识别算法。首先,利用迭代加权低秩分解算法分别提取各类训练样本中包含的遮挡和噪声因素。然后针对测试样本和训练样本遮挡情况有差异的问题,利用迭代加权低秩分解算法提取测试样本中包含遮挡所掩盖的信息。最后将每类训练样本的低秩矩阵、遮挡矩阵、噪声矩阵和测试样本中的遮挡向量构造新的联合字典,将测试样本表示为新的联合字典的稀疏线性组合,利用稀疏逼近计算残差,通过得到的系数进行分类判别。实验结果表明,基于迭代加权低秩分解的遮挡人脸识别算法在AR和ExtendedYaleB库上的识别率得到提高,相比其他方法有较好的识别结果,验证了该算法的有效性,对于遮挡情况具有很好的鲁棒性。

Aiming at the problem that the sparse matrix obtained by the traditional low rank matrix decomposition algorithm contains both occlusion factors and noise factors,we propose an occlusion face recognition algorithm based on iteratively reweighted robust principal component analysis. First,the iteratively reweighted robust principal component analysis is used to extract the occlusion and noise factors contained in each training sample. Then,for the problem that the test sample and the training sample have different occlusion conditions,the iteratively reweighted robust principal component analysis is used to extract the information covered by the occlusion contained in the test sample. Finally,low-rank matrix,occlusion matrix,noise matrix and occlusion vector in the test sample of each type of training samples are constructed into a new joint dictionary,and the test samples are represented as sparse linear combinations of the new joint dictionary,and the residuals are calculated by sparse approximation. The classification is determined by the obtained coefficients. The experiment shows that the recognition rate of the proposed algorithm is improved on AR and Extended Yale B library. Compared with other methods,it has better recognition results,which is proved to be effect,and is robust to occlusion.