Float greedy-search-based subspace clustering

Many kinds of efficient greedy subspace clustering methods have been proposed to cut down the computation time in clustering large-scale multimedia datasets. However, these methods are easy to fall into local optimum due to the inherent characteristic of greedy algorithms, which are step-optimal only. To alleviate this problem, this paper proposes a novel greedy subspace clustering strategy based on floating search, called Float Greedy Subspace Clustering (FloatGSC). In order to control the complexity, the nearest subspace neighbor is added in a greedy way, and the subspace is updated by adding an orthogonal basis involved with the newly added data points in each iteration. Besides, a backtracking mechanism is introduced after each iteration to reject wrong neighbors selected in previous iterations. Extensive experiments on motion segmentation and face clustering show that our algorithm can significantly improve the clustering accuracy without sacrificing much computational time, compared with previous greedy subspace clustering methods.