Layout consistent segmentation of 3-D meshes via conditional random fields and spatial ordering constraints.

Area of Science:

• Computer Vision
• Geometric Modeling
• Machine Learning

Background:

• 3-D mesh segmentation is crucial for understanding object structure.
• Existing methods struggle with objects having known part structures and semantic consistency.

Purpose of the Study:

• To develop a robust 3-D mesh segmentation technique for objects with known part structures.
• To ensure semantic consistency in part labeling through spatial arrangement constraints.

Main Methods:

• Utilized Conditional Random Fields (CRF) to model label distribution and enforce spatial constraints.
• Incorporated intrinsic surface shape descriptors for each label.
• Employed randomized decision trees for model learning and graph cuts for segmentation.

Main Results:

• Achieved accurate semantic interpretation of non-overlapping subsurfaces.
• Demonstrated robustness to local and global shape variations.
• Successfully segmented geometrically less structured regions.

Conclusions:

• The proposed CRF-based approach effectively segments 3-D meshes with known part structures.
• The method ensures semantic consistency and handles shape variations, offering a flexible solution.