自适应提升的正交匹配寻踪稀疏子空间聚类

Sparse subspace clustering algorithm has attracted much attention due to its broad theoretical guarantees and excellent experimental performance. However, the self-expression coefficients calculated by the sparse subspace clustering algorithm is often too sparse to ensure that all data points from the same subspace in the induced affinity graph constitute connected components --- which will cause serious over-segmentation problems in subsequent spectral clustering, leading to a significant drop in clustering accuracy. This paper attempts to introduce an ensemble learning idea into sparse subspace clustering algorithms to increase non-zero self-expression coefficients. Specifically, the sparse self-expression coefficients are solved via Orthogonal Matching Pursuit algorithm multiple times, and the data points associated with non-zero self-expression coefficients are automatically suppressed when solving the self-expression coefficients in the next time; therefore, it is called Adaptive Boosted Sparse Subspace Clustering via Orthogonal Matching Pursuit (AdaBoost SSCOMP). In AdaBoost SSCOMP, the computed self-expression coefficients are integrated and then induce a connectivity improved affinity graph. To validate the effectiveness of the proposed approach, extensive experiments are conducted on both synthetic and real world datasets, yielding notable performance improvement.