论文部分内容阅读
针对现有子空间聚类算法在构造相似度矩阵时，没有同时利用样本自表达和稀疏相似度矩阵以及去除噪音、离群点的干扰相结合，提出了一种新的稀疏样本自表达子空间聚类方法。该方法通过样本自表达而充分利用样本间固有相关性的本质，创新性地同时使用L1?范数和L2，1?范数正则化项惩罚相似度矩阵，即对所有测试样本进行稀疏样本自表达，从而确保每个测试样本由与其相关性强的样本表示，并使所获得的相似度矩阵具有良好的子空间结构和鲁棒性。通过Hopkins155和人脸图像等大量数据集的实验结果表明，本文方法在实际数据的子空间聚类中能够获得非常好的效果。“,”Existing subspace clustering methods do not combine sample self?representation well with affinity matrix sparsity, for example, by removing disturbances from noise, outliers, etc., when constructing the affinity matrix. This paper proposes a novel subspace clustering method called sparse sample self?representation for subspace cluste?ring. This method fully considers the correlation between the samples, and also takes advantage of L1?norm and L2,1?norm terms to “penalize” the affinity matrix;that is, it conducts sparse sample self?representation for all test samples, to guarantee every sample can be expressed by any other samples with strong similarity and make it more robust. The experimental results of the Hopkins155 dataset and some facial image datasets show that the proposed method outperforms the LSR, SSC, and LRR methods in terms of the subspace clustering error.