Active testing search for point cloud matching.

Area of Science:

• Computational geometry
• Medical image analysis
• Computer vision

Background:

• Point-cloud matching is crucial for aligning 3D data in various scientific fields.
• Existing methods often struggle with mildly nonlinear transformations and require initial alignment.
• Accurate registration of graph structures, such as neuronal networks, presents unique challenges.

Purpose of the Study:

• To develop a general and robust approach for point-cloud matching under mildly nonlinear transformations.
• To enable accurate registration of graph structures using geometric information only.
• To provide a faster and more comprehensive solution compared to existing methods.

Main Methods:

• A two-stage approach combining Active Testing Search (ATS) for coarse approximation and dynamic programming for refinement.
• ATS explores a reduced set of partial matches based on geometric point positions.
• Branching point matching is used for graph structure registration without prior alignment knowledge.

Main Results:

• The proposed method successfully registered point clouds with mildly nonlinear transformations.
• Demonstrated effectiveness on diverse datasets including angiography, retinal fundus, and neuronal microscopy data.
• Achieved superior performance in cases where other methods failed, with increased speed.

Conclusions:

• The developed point-cloud matching algorithm offers a robust and efficient solution for complex registration tasks.
• Its geometric-based approach and two-stage refinement make it broadly applicable to scientific data.
• The method advances the state-of-the-art in point-cloud registration, particularly for graph structures.