A skeleton family generator via physics-based deformable models.

Area of Science:

• Computer Vision
• Image Analysis
• Computational Geometry

Background:

• Skeleton extraction is crucial for shape analysis and object recognition.
• Existing methods often struggle with noise, shape variations, and topological accuracy.
• A need exists for robust skeletonization techniques that preserve object topology.

Purpose of the Study:

• To present a novel 2-D physics-based deformable model for object skeleton family extraction.
• To demonstrate the generation of a diverse family of skeletons from a single object representation.
• To validate the accuracy and robustness of the proposed skeletonization approach.

Main Methods:

• Utilized a 2-D physics-based deformable model to parameterize object shape.
• Solved deformation equations using modal analysis.
• Generated skeleton families by varying model physical characteristics.

Main Results:

• Obtained skeletons closely match hand-labeled skeletons from human subjects.
• The method is robust to significant noise, shape variations, cuts, and tears.
• Generated skeletons preserve the original object's topology without spurious branches.
• No post-processing skeleton pruning methods were required.

Conclusions:

• The proposed physics-based deformable model offers a robust and accurate method for object skeleton family extraction.
• The technique successfully generates topologically consistent skeletons, outperforming traditional methods in challenging conditions.
• This approach advances skeletonization by producing families of skeletons without relying on pruning algorithms.