Approximately global optimization for robust alignment of generalized shapes.

Area of Science:

• Computer Vision
• Computational Geometry
• Image Processing

Background:

• Shape alignment is crucial in computer vision and graphics.
• Existing methods often struggle with diverse shape representations and computational efficiency.
• Robust and efficient shape dissimilarity metrics are needed for global optimization.

Purpose of the Study:

• To introduce a novel, flexible image-based shape representation.
• To develop a robust and efficient energy function for shape dissimilarity.
• To adapt and enhance Particle Swarm Optimization for accurate global shape alignment.

Main Methods:

• Utilizing gray-scale images for source shapes.
• Proposing a two-component Gaussian Mixture (GM) distance map for target shapes.
• Developing a new energy function based on the GM distance model.
• Modifying Particle Swarm Optimization (PSO) with enhanced strategies for robust global optimization.

Main Results:

• The proposed GM distance map representation effectively handles various shape data types.
• The new energy function serves as a robust and efficient shape dissimilarity metric.
• The modified PSO demonstrates improved robustness and avoids premature convergence.
• The framework shows superior performance compared to state-of-the-art methods on 2D and 3D data.

Conclusions:

• The novel method provides a robust and efficient solution for shape alignment problems.
• The asymmetric GM distance map representation offers flexibility for diverse shape data.
• The enhanced PSO effectively optimizes the proposed energy function for accurate results.