Latent Distribution Preserving Deep Subspace Clustering

Abstract

Subspace clustering is a useful technique for many computer vision applications in which the intrinsic dimension of high-dimensional data is smaller than the ambient dimension. Traditional subspace clustering methods often rely on the self-expressiveness property, which has proven effective for linear subspace clustering. However, they perform unsatisfactorily on real data with complex nonlinear subspaces. More recently, deep autoencoder based subspace clustering methods have achieved success owning to the more powerful representation extracted by the autoencoder network. Unfortunately, these methods only considering the reconstruction of original input data can hardly guarantee the latent representation for the data distributed in subspaces, which inevitably limits the performance in practice. In this paper, we propose a novel deep subspace clustering method based on a latent distribution-preserving autoencoder, which introduces a distribution consistency loss to guide the learning of distribution-preserving latent representation, and consequently enables strong capacity of characterizing the real-world data for subspace clustering. Experimental results on several public databases show that our method achieves significant improvement compared with the state-of-the-art subspace clustering methods.

Deep Closed-Form Subspace Clustering

“…It is also noteworthy that performance of the DCFSC in this experiment was reported to be comparable with 26.6%, which was reported in the study [37] combining more sophisticated methodologies such as self-supervised learning with the DSC. Since DCFSC is easy to combine with the more advanced models [41,37,40,39] of DSC, there is possibility of further enhancing the performance of these modified models with deeper neural architecture. 3…”

“…The biggest contribution of [12] was that they firstly designed the self-expressive layer and corresponding loss function which models self-expressiveness property of data into deep auto-encoder. Since DSC showed great performance on various benchmarks, there have been many subsequent studies [41,37,40,39] that tried to improve the DSC in several aspects. Deep adversarial subspace clustering [41] exploited GAN-like adversarial learning framework to supervise sample representation learning and subspace clustering.…”

Self-Supervised Deep Subspace Clustering for Hyperspectral Images With Adaptive Self-Expressive Coefficient Matrix Initialization

“…2) Subspace Clustering in Feature Space: This type of methods [17]- [19], [35] map the raw samples into feature space to better capture the non-linear characteristics of sample distribution, and then construct the affinity matrix in the feature space. For instance, kernel subspace clustering [17] is proposed to implicitly map the HSI samples from original space to a kernelized space.…”

Subspace Clustering and Feature Typicality Degrees: a Prospective Study

“…SSC [8], [9] or LRR [21]). Deep learning methods have been proposed to extend this principle to the case of non-linear subspaces [22], [14], [27].…”