The elastic ratio: introducing curvature into ratio-based image segmentation.

Area of Science:

• Computer Vision
• Image Processing
• Computational Geometry

Background:

• Existing image segmentation methods often require user input or lack control over region boundary characteristics.
• Ratio-based frameworks offer a promising avenue but have limitations in incorporating geometric properties like curvature.

Purpose of the Study:

• To develop a novel ratio-based image segmentation method that integrates curvature regularity for enhanced boundary definition.
• To generalize existing ratio frameworks to accommodate curvature penalties.
• To provide a fully unsupervised segmentation approach with guaranteed convergence.

Main Methods:

• The study casts image segmentation as finding minimal ratio cyclic paths in a graph where nodes represent line segments.
• A generalized ratio framework is proposed, focusing on the elastic ratio which balances region size and curvature.
• The method utilizes a graph-based algorithm for discrete minimization and proves convergence to continuous solutions.

Main Results:

• The proposed method successfully incorporates curvature regularity, leading to substantial improvements in segmentation quality.
• Numerical experiments demonstrate the effectiveness of the unsupervised, curvature-aware approach.
• The algorithm identifies meaningful global optima for a parameterization-independent snakes functional.

Conclusions:

• The developed ratio-based segmentation method with curvature regularity offers superior, unsupervised segmentation results.
• The approach provides a robust solution for defining region boundaries with controlled smoothness.
• This work advances unsupervised image segmentation by integrating geometric constraints effectively.