Elastic Net Constraint-Based Tensor Model for High-Order Graph Matching

Area of Science:

• Computer Vision
• Machine Learning
• Optimization

Background:

• Establishing correspondences between feature points is crucial for computer vision tasks.
• Existing graph matching methods face challenges in balancing result sparsity and accuracy.

Purpose of the Study:

• To propose a novel elastic net constraint-based tensor model for high-order graph matching.
• To enhance the trade-off between sparsity and accuracy in matching results.

Main Methods:

• An elastic net constraint was integrated into a tensor-based graph matching model.
• A nonmonotone spectral projected gradient (NSPG) method was developed for optimization.
• A projection algorithm for elastic net constraints was proposed.

Main Results:

• The proposed model effectively controls the trade-off between sparsity and accuracy.
• The NSPG method demonstrated global convergence for solving the matching model.
• Experimental results on synthetic and natural images validated the method's superiority.

Conclusions:

• The elastic net constraint-based tensor model offers an effective solution for high-order graph matching.
• The NSPG optimization method provides a robust and convergent approach.
• The proposed method advances computer vision applications requiring accurate feature point correspondence.