Abstract:Multi-view subspace clustering has received widespread attention. Since data may violate the linear assumption in many practical applications, multi-view subspace clustering methods based on deep neural network are developed in recent years. However, most existing DNN based methods only focus on the utilization of the deepest features (i.e., features extracted from the deepest layer), regardless of the informative shallow features (i.e., features extracted from the shallow ones) and the fusion of multi-scale features (i.e., deep and shallow features), which may make their clustering performance suboptimal. In addition, these methods usually do not impose appropriate constraints on the extracted features to ensure that they can maintain the inherent local geometric structure of multi-view data, which is unfavorable for subspace learning. To deal with the limitations mentioned above, we propose a novel DNN based subspace clustering method for multi-view data called Deep Multi-view Subspace Clustering via Structure-preserved Multi-scale Features Fusion. Specifically, the graph knowledge of multi-view data is introduced to constrain the extracted features so that they have the structure-preserved property. Simultaneously, by fully exploiting the consistency of the extracted structure-preserved multi-scale features, DMSC-SMFF can learn the more discriminative common shared subspace representation, which can be employed by the spectral clustering module to obtain clustering results. Moreover, the optimization algorithm based on the Alternating Direction Method (ADM) is developed to optimize our objective function. Furthermore, DMSC-SMFF can also be applied to single-view data. Plentiful experimental results on five benchmark datasets demonstrate the effectiveness and superiority of the proposed method.

Incomplete Multi-View Subspace Clustering with Adaptive Instance-Sample Mapping and Deep Feature Fusion

Multi-View Clustering Via Simultaneously Learning Shared Subspace And Affinity Matrix

Incomplete Multi-View Clustering Via Deep Semantic Mapping

Adaptive Weighted Graph Fusion Incomplete Multi-View Subspace Clustering

Deep Incomplete Multi-View Clustering Via Mining Cluster Complementarity.

Deep Multi-View Subspace Clustering Via Structure-Preserved Multi-Scale Features Fusion

Multi-scale deep multi-view subspace clustering with self-weighting fusion and structure preserving

View-interactive attention information alignment-guided fusion for incomplete multi-view clustering

Adaptive Feature Projection With Distribution Alignment for Deep Incomplete Multi-View Clustering

Incomplete multi-view subspace clustering based on missing-sample recovering and structural information learning

Dynamic Incomplete Multi-view Imputing and Clustering

Multi‐view Subspace Clustering with Incomplete Graph Information

Adaptive Feature Imputation with Latent Graph for Deep Incomplete Multi-View Clustering

Incomplete Multiview Clustering Via Late Fusion

Incomplete Multi-View Clustering with Reconstructed Views

Incomplete Multi-View Clustering Via Inference and Evaluation

Self-Representation Subspace Clustering for Incomplete Multi-view Data.

Efficient and Effective Deep Multi-view Subspace Clustering

Adversarial Incomplete Multiview Subspace Clustering Networks

Deep Multi-View Clustering via Multiple Embedding.

Incomplete multi-view clustering based on weighted sparse and low rank representation