Multiple Kernel Sparse Representation-Based Orthogonal Discriminative Projection and Its Cost-Sensitive Extension.

Area of Science:

• Computer Science
• Machine Learning
• Pattern Recognition

Background:

• Sparse Representation-based Classification (SRC) is effective but limited by linearity.
• Existing methods like SRC-DP are linear, while KSRC requires difficult kernel selection.
• Multiple Kernel Learning for SRC (MKL-SRC) overlooks between-class information during kernel learning.

Purpose of the Study:

• To propose a novel multiple kernel sparse representation classifier.
• To develop a method that learns projection matrices and kernels to optimize discrimination.
• To incorporate cost information for improved recognition in low-dimensional subspaces.

Main Methods:

• A novel multiple kernel sparse representation-based classifier is proposed.
• A multiple kernel sparse representation-based orthogonal discriminative projection method is designed.
• Trace ratio optimization is used to efficiently solve for projection matrices and kernels.

Main Results:

• The proposed algorithm effectively maximizes between-class reconstruction residuals and minimizes within-class residuals.
• Incorporating cost information further enhances recognition performance in the learned subspace.
• Experimental results show the superiority of the proposed method over state-of-the-art techniques.

Conclusions:

• The novel multiple kernel sparse representation classifier offers significant improvements over existing methods.
• The approach effectively handles nonlinear data distributions and optimizes discriminative projections.
• The method demonstrates superior performance in classification tasks, particularly in low-dimensional subspaces.