Affine non-negative collaborative representation based pattern classification

Area of Science:

• Computer Science
• Machine Learning
• Pattern Recognition

Background:

• Representation-based classification is crucial in pattern recognition.
• Non-negative representation based classification (NRC) shows promise but has limitations.
• NRC lacks regularization and doesn't account for data residing in multiple affine subspaces.

Purpose of the Study:

• To introduce an improved pattern classification model called affine non-negative collaborative representation (ANCR).
• To address the drawbacks of NRC, specifically the lack of regularization and the handling of affine subspaces.
• To enhance classification accuracy and stability in pattern recognition tasks.

Main Methods:

• Developed the affine non-negative collaborative representation (ANCR) model.
• Incorporated a regularization term into the coding vector formulation.
• Introduced an affine constraint to better represent data within affine subspaces.

Main Results:

• ANCR demonstrated superior performance compared to NRC on benchmarking datasets.
• Achieved 97.8% accuracy on the Hopkins dataset and 87.7% on the Aircraft dataset.
• Showcased improvements of 2.2% and 0.4% over NRC, respectively.

Conclusions:

• The proposed ANCR model effectively enhances pattern classification.
• The integration of regularization and affine constraints leads to more stable and accurate results.
• ANCR offers a significant advancement over existing non-negative representation based classification methods.