Self-supervised deep subspace clustering with entropy-norm

Auto-Encoder based Deep Subspace Clustering (DSC) has been widely applied in computer vision, motion segmentation and image processing. However, existing DSC methods suffer from two limitations: (1) they ignore the rich useful relational information connectivity within each subspace due to reconstruction loss; (2) design convolutional networks individually according specific datasets. To address above problems improve performance of DSC, we propose a novel algorithm called Self-Supervised deep with Entropy-norm(S $$^{3}$$ CE) this paper. Firstly, S CE introduces self-supervised contrastive learning pre-train encoder instead requiring decoder. Besides, trained is used as feature extractor segment by combining self-expression layer entropy-norm constraint. This not only preserves local structure data, but also improves between data points. Extensive experimental results demonstrate superior comparison state-of-the-art approaches.