基于干净字典学习的多视图聚类算法

Multi-View Clustering Algorithm Basedon Clean Dictionary Learning

摘要: 多视图聚类已被广泛应用到图像分类、信息检索、医学病理分析等多个领域。虽然基于自表示学 习的多视图子空间聚类算法在学术界已有诸多研究，但这些方法大多数是直接利用原始数据作为 字典来构建亲和图，因此，聚类性能往往会受到原始数据特征质量的影响。为了克服这一问题， 我们提出了一种全新的多视图聚类算法，名为 “基于干净字典学习的多视图聚类算法”。在这种 方法中，我们首先将原始数据分解为干净数据和噪声数据，然后通过采用干净数据矩阵来进行字 典学习，从而避免了直接使用受污染数据所导致的错误表示。同时，我们引入了鲁棒主成分分析 (RPCA) 和秩约束，以构建出更加干净、鲁棒的亲和矩阵。最后，我们使用了基于增广拉格朗日 乘子 (ALM) 的优化方法来求解模型的目标函数。实验结果表明，在 4 个真实的多视图数据集上， 我们算法的聚类性能均优于其他先进的聚类算法，展现出了非凡的性能。

Abstract: Multi-view clustering has been widely used in many fields such as image classification, information retrieval, and medical pathology analysis. Although multi-view subspace clustering algorithms based on self-representation learning have been researched in many academic circles, most of these methods directly utilize the original data as a dictionary to construct affinity graphs, and thus the clustering performance tends to be affected by the quality of features in the original data. To overcome this problem, we propose a novel multi-view clustering algorithm called “Multi-view clustering algorithm based on clean dictionary learning”. In this approach, we first decompose the original data into clean and noisy data, and then perform dictionary learning by employing a clean data matrix, which avoids the misrepresentation caused by the direct use of contaminated data. At the same time, we introduce Robust Principal Component Analysis (RPCA) and rank constraints to construct a cleaner and more robust affinity matrix. Finally, we use an optimization method based on augmented Lagrange multipliers (ALM) to solve the objective function of the model. Experimental results show that the clustering performance of our algorithm outperforms other state-of-the-art clustering algorithms on all four real multi-view datasets, demonstrating exceptional performance.