Kernel Two-Dimensional Ridge Regression for Subspace Clustering

TL;DR

This paper introduces a novel 2D subspace clustering method that directly processes 2D data, preserving inherent structures and relationships, and jointly learns projections and representations for improved clustering performance.

Contribution

It proposes a new kernel two-dimensional ridge regression approach for subspace clustering that directly handles 2D data, unlike existing methods that convert data into vectors.

Findings

Effective in preserving data structures

Outperforms existing clustering methods

Converges reliably with an efficient algorithm

Abstract

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.