Segmentation of 3D Meshes Using p-Spectral Clustering

Area of Science:

• Computer Vision
• Computational Geometry
• Cognitive Science

Background:

• 3D mesh segmentation is crucial for various applications.
• Existing methods often lack human-like perception or require supervision.
• Hierarchical and perceptually relevant segmentation remains a challenge.

Purpose of the Study:

• To develop a fully unsupervised method for 3D mesh segmentation.
• To achieve segmentation that aligns with human perception.
• To provide a hierarchical segmentation of 3D meshes.

Main Methods:

• Utilizing the minima rule and spectral clustering.
• Developing a novel adjacency matrix based on cognitive studies.
• Employing one-spectral clustering for optimal Cheeger cut values.

Main Results:

• The proposed method achieves optimal 3D mesh segmentation.
• The segmentation is hierarchical and unsupervised.
• The approach is grounded in cognitive principles for human-like perception.

Conclusions:

• The new approach effectively segments 3D meshes in a human-perceivable manner.
• Unsupervised learning combined with cognitive insights offers a powerful tool for mesh analysis.
• This method advances the field of 3D mesh processing.