Sparse Subspace Learning Based on Learnable Constraints for Image Clustering

Sparse subspace clustering is a widely used method for clustering high dimensional data, but the traditional method is complex and requires prior information that can be difficult to obtain in unsupervised scenarios. In this paper, we propose a new method called Self-constrained Sparse Subspace Clustering (ScSSC) that adds two self-constraints to find prior information, simplifying the clustering of high dimensional data. The proposed algorithm is a non-deep neural network model that extends the traditional sparse subspace clustering objective function and transforms the clustering problem into a spectral clustering optimization problem. The algorithm can discover a high-quality cluster structure without prior information, making it highly effective in unsupervised scenarios. Our experiment analysis shows that the proposed algorithm outperforms other comparison methods in terms of three metrics. The algorithm's robustness and stability are further demonstrated through ablation experiments and parameter analysis. The proposed algorithm reduces the complexity of the clustering method, making it a valuable tool in understanding and analyzing information in datasets.

Automated summary

Sparse subspace clustering is a widely used method for clustering high dimensional data. However, the traditional method is complex and requires prior information. In this paper, we propose a new method called Self-constrained Sparse Subspace Clustering (ScSSC) to simplify the clustering of high dimensional data. The proposed algorithm is a non-deep neural network model that can discover a high-quality cluster structure without prior information, making it highly effective in unsupervised scenarios.