关键词: 谱聚类, 局部约束, 多准则融合, 图构建

Abstract:

Spectral clustering is an effective subspace clustering method which is widely used in image clustering, image segmentation and others. However, the performance of spectral clustering depends on the construction of graph to a certain extent. Therefore, how to construct an effective graph becomes the key problem in spectral clustering. For the shortcomings of existing graph construction methods, this paper proposes a new method named Multiple Locality Constrained Self Representation (MLCSR) for graph construction. This proposed method considers not only the self-representation ability of the samples, but also the local structure information of the samples. Particularly, it exploits the adaptive weighting to fuse different distance measures when building local constraints. In addition, an iterative optimization algorithm is designed to solve the objective function. Finally, the effectiveness of the proposed method is verified on three standard face databases.

Key words: spectral clustering, locality constraint, multi-criteria fusion, graph construction