基于k近邻的快速和参数自适应训练的稀疏子空间聚类

Fast and parameter adaptive training sparse subspace clustering based on k-nearest neighbor

稀疏子空间聚类算法具有较强的子空间识别能力和灵活的建模特性,但该算法存在复杂度高、参数敏感及聚类结果不稳定等问题。对此,本文提出了一种将高效率近邻过滤和参数自适应训练相结合,并应用于稀疏子空间聚类模型的算法。该算法通过k近邻算法筛选重构样本的候选点,并利用数据全局关系自适应地拟合正则参数,改变了原始稀疏子空间聚类自表示数据点和正则参数的选取方式。通过仿真实验验证了提出的算法不仅降低了运算成本,而且能够自适应选择参数,提高了聚类精度。

Sparse subspace clustering algorithm has strong subspace recognition ability and flexible modeling characteristics,but it also has some problems such as high complexity,sensitive parameters and unstable clustering results.In this paper,an algorithm combining efficient neighbor filtering and parameter adaptive training is proposed and applied to sparse subspace clustering model.This algorithm selects the candidate points of reconstructed samples by k-nearest neighbor algorithm,and adaptively fits the canonical parameters by using the global relation of data,which changes the selection method of the original sparse subspace clustering self-representation data points and canonical parameters.Simulation results show that the proposed algorithm not only reduces the operation cost,but also adaptively selects parameters and improves the clustering accuracy.