Subspace clustering applied to face images

Abstract

In this paper, two state-of-the-art subspace clustering techniques, namely the Sparse Subspace Clustering and the Elastic Net Subspace Clustering, are tested for clustering. Both algorithms are frequently implemented using the linearized alternating directions method. An efficient implementation of the Elastic Net Subspace Clustering is derived, employing the fast iterative shrinkage algorithm. Random projections are also used to reduce significantly the computation time. Figures of merit are reported for two publicly available face image datasets, i.e., the Extended Yale B dataset and the Hollywood dataset.