Kernel Two-Dimensional Ridge Regression for Subspace Clustering
This addresses the issue of structure loss in subspace clustering for 2D data, which is incremental as it builds on existing methods by avoiding vectorization.
The paper tackles the problem of subspace clustering for 2D data by proposing a method that directly uses 2D inputs to preserve inherent structures, resulting in improved performance verified through extensive experiments.
Subspace clustering methods have been widely studied recently. When the inputs are 2-dimensional (2D) data, existing subspace clustering methods usually convert them into vectors, which severely damages inherent structures and relationships from original data. In this paper, we propose a novel subspace clustering method for 2D data. It directly uses 2D data as inputs such that the learning of representations benefits from inherent structures and relationships of the data. It simultaneously seeks image projection and representation coefficients such that they mutually enhance each other and lead to powerful data representations. An efficient algorithm is developed to solve the proposed objective function with provable decreasing and convergence property. Extensive experimental results verify the effectiveness of the new method.