Optimal Couple Projections for Domain Adaptive Sparse Representation-Based Classification.

Area of Science:

• Machine Learning
• Computer Vision
• Pattern Recognition

Background:

• Sparse Representation-based Classification (SRC) is a successful classification method.
• SRC performance degrades significantly when training and testing data distributions differ.
• Domain shift presents a major challenge for existing SRC methods.

Purpose of the Study:

• To propose an optimal couple projections for domain-adaptive SRC (OCPD-SRC) method.
• To address the performance degradation of SRC under domain shift.
• To learn discriminative features and a common dictionary for cross-domain classification.

Main Methods:

• OCPD-SRC simultaneously learns coupled projection matrices and a common discriminative dictionary.
• The method maximizes between-class sparse reconstruction residuals and minimizes within-class residuals.
• An alternative optimization method is used to efficiently obtain the optimal solution.

Main Results:

• The learned representations fit SRC and possess enhanced discriminant ability.
• OCPD-SRC demonstrates superior or comparable performance to state-of-the-art methods on benchmark databases.
• The method is extendable to multiple domains and can handle nonlinear data structures via kernelization.

Conclusions:

• OCPD-SRC effectively overcomes the domain shift problem in sparse representation-based classification.
• The proposed method offers a robust solution for cross-domain classification tasks.
• OCPD-SRC provides a flexible framework adaptable to various data complexities and domain numbers.