Block-Diagonal Constrained Low-Rank and Sparse Graph for Discriminant Analysis of Image Data

Area of Science:

• Machine Learning
• Computer Vision
• Data Science

Background:

• Dimensionality reduction (DR) is crucial for analyzing high-dimensional data.
• Low-rank and sparse model-based DR methods are gaining attention.
• Existing methods may not fully capture both local and global data structures or ensure class separability.

Purpose of the Study:

• To propose an effective supervised dimensionality reduction technique named Block-Diagonal Constrained Low-Rank and Sparse-based Embedding (BLSE).
• To reveal intrinsic intra-class and inter-class relationships, local neighborhood relations, and global data structure.
• To enhance discrimination between different classes in the embedded subspace.

Main Methods:

• BLSE comprises two steps: Block-Diagonal Constrained Low-Rank and Sparse Representation (BLSR) and Block-Diagonal Constrained Low-Rank and Sparse Graph Embedding (BLSGE).
• BLSR incorporates sparse constraints for local structure, low-rank criteria for global structure, and block-diagonal regularization for class discrimination.
• BLSGE constructs informative graphs and seeks a low-dimensional embedding by minimizing intra-class scatter and maximizing inter-class scatter.

Main Results:

• Experiments on benchmark face and object image datasets demonstrated the effectiveness of the proposed BLSE approach.
• The method successfully revealed local and global data structures.
• Enhanced class discrimination was observed in the low-dimensional embedding.

Conclusions:

• BLSE is an effective supervised dimensionality reduction technique.
• The proposed method offers improved feature representation by considering local, global, and inter-class structures.
• BLSE shows promise for applications in image recognition and other high-dimensional data analysis tasks.