Minimum near-convex shape decomposition.

Area of Science:

• Computer Vision
• Computational Geometry
• Shape Analysis

Background:

• Part-based shape representation is crucial for understanding complex objects.
• Existing decomposition methods often produce excessive or unnatural parts.
• There is a need for robust and perceptually guided shape decomposition techniques.

Purpose of the Study:

• To develop a novel method for decomposing arbitrary shapes into a minimum number of near-convex parts.
• To enhance the visual naturalness and robustness of shape decomposition.
• To provide an efficient optimization framework for shape decomposition.

Main Methods:

• Formulated near-convex shape decomposition as a discrete optimization problem, minimizing nonintersecting cuts.
• Incorporated two perception rules as constraints to improve visual naturalness.
• Utilized binary integer linear programming for efficient optimization.
• Introduced a user-specified parameter for the degree of near-convexity.

Main Results:

• The proposed minimum near-convex decomposition (MNCD) effectively decomposes shapes into a minimal set of near-convex parts.
• The method demonstrates robustness to local distortions and shape deformations.
• MNCD outperforms state-of-the-art methods in terms of part redundancy and representation quality.
• Theoretical analysis and experimental results validate the approach's effectiveness.

Conclusions:

• MNCD provides a robust and perceptually relevant approach to shape decomposition.
• The method achieves efficient and accurate decomposition without redundant parts.
• This work advances part-based shape representation through optimized, near-convex decomposition.