Abstract:Spectral clustering is a recent proposed popular method for unsupervised image clustering problem, whose performance relies on a well-designed affinity matrix that measures the similarity among data. Nevertheless, on one hand, most of the existed affinity learning methods have been proposed focus on single-feature data, which would lose some cues from their other useful heterogeneous features. On the other hand, the specific two dimensional structures of visual data are not well preserved when affinity is learned. To address these two problems, in this paper, we propose a novel heterogeneous feature based strategy for image clustering by combining projection learning and affinity learning into a joint learning framework. Inspired by multimodal information integration, heterogeneous modal-based projection learning scheme, that maps each type of heterogeneous feature into the latent subspace in a two-dimensional way, is built with original image and their multi-modal heterogeneous features respectively to preserve more useful structural information. Then, the ensembling projected data of each modal are jointly used to establish an affinity learning model based on the shared ridge regression that considers cues from all modals simultaneously, which can greatly improve the similarity metric performance. Moreover, the graph-based constraint on affinity matrix is also adopt to capture the extra nonlinear structures of the data, which provides help for keeping their neighbor relationships. An iterative numerical scheme is designed to solve the objective function of our proposed framework and ensure convergence. Extensive experimental results on several public datasets verify the advancement and effectiveness of our proposed method among comparison methods.

Structure-Aware Preserving Projections with Applications to Medical Image Clustering

Structure-preserving visualization for single-cell RNA-Seq profiles using deep manifold transformation with batch-correction

Structure Preserving Projections Learning Via Low-Rank Embedding for Image Classification

Structure-Aware Subspace Clustering

Orthogonal Self-Guided Similarity Preserving Projection for Classification and Clustering.

Linear Dimensionality Reduction Based on Hybrid Structure Preserving Projections

Local Feature Discriminant Projection

Category Guided Sparse Preserving Projection for Biometric Data Dimensionality Reduction.

Lifting the curse from high dimensional data: Automated projection pursuit clustering for the variety of biological data modalities

Local and Global Preserving Semisupervised Dimensionality Reduction Based on Random Subspace for Cancer Classification

Understanding How Dimension Reduction Tools Work: An Empirical Approach to Deciphering t-SNE, UMAP, TriMAP, and PaCMAP for Data Visualization

A Dimension Reduction Algorithm Preserving Both Global and Local Clustering Structure

Locality Preserving Projection with Symmetric Graph Embedding for Unsupervised Dimensionality Reduction.

Semi-supervised Enhanced Discriminative Local Constraint Preserving Projection for Dimensionality Reduction of Medical Hyperspectral Images

Clustering Preserving Projections for High-Dimensional Data

A Scalable Unsupervised Feature Merging Approach to Efficient Dimensionality Reduction of High-Dimensional Visual Data

A projective clustering algorithm based on significant local dense areas

Multi-view Sparsity Preserving Projection for dimension reduction.

Supervised Discriminative Sparse PCA with Adaptive Neighbors for Dimensionality Reduction

Multi-view Subspace Clustering via Two Dimensional Structures Preservation on Heterogeneous Features

Revisiting Dimensionality Reduction Techniques for Visual Cluster Analysis: An Empirical Study