A Topology fixated Shape Gradient Framework for Non Simple Boundary Extraction for CIE Lab color images with Repulsive Energy
arXiv cs.CV / 4/28/2026
📰 NewsIdeas & Deep AnalysisModels & Research
Key Points
- The paper proposes a hybrid, level-set-free image segmentation framework for CIE Lab color images using a modified piecewise-constant shape gradient derived from a Mumford–Shah functional combined with a repulsive energy term.
- Segmentation is carried out via non-local evolution of discrete curves, enabling the method to handle images with disjoint regions and multiple boundaries.
- A novel multivariable component, dependent on a small set of sampled points on the evolving curves, is introduced to prevent or manage self-intersections during boundary evolution.
- Experiments on grayscale and color images—including nested structures and astronomical targets—show effective segmentation with strong control over segment topology and boundary self-intersections.
- The framework emphasizes precise topological control, suggesting improved robustness for complex segmentation scenarios where standard boundary evolution can fail.
