Review on selective variational image segmentation: models, geometric constraints, and convex reformulations

Mohamed Shariff, Mohamed Ammareza and Khairunnizam, Nuruljannah and Jumaat, Abdul Kadir (2026) Review on selective variational image segmentation: models, geometric constraints, and convex reformulations. Mathematics Letters, 5 (2): 12. pp. 136-145. ISSN eISSN: 2948-3735
Abstract

This comprehensive review explores the evolution of selective and interactive variational image segmentation, organizing the current literature around three core axes: energy formulations, geometric constraints, and convex reformulations. While traditional global segmentation methods partition entire images based on statistical properties, they often struggle with noise, poor contrast, and intensity inhomogeneity. In contrast, selective models utilize user-supplied markers to construct geometric constraints, effectively anchoring the evolving active contour to specific target objects rather than every homogeneous region. We examine how these marker-driven terms are integrated into two-phase energy functionals, balancing data fidelity with boundary regularization to prevent contour leakage into adjacent structures. A primary focus of this review is the shift towards convex reformulations. By relaxing non-convex level-set energies, these models achieve initialization-independent global minimizers that can be efficiently solved using primal-dual algorithms. Finally, we discuss critical open challenges, emphasizing the need for automated marker generation via learned shape priors and the rigorous validation of convex selective frameworks on vector-valued and multi-modal medical imaging datasets.

Item Details
Edit Item
Edit Item
Downloads & Files
[thumbnail of 150077.pdf]
Text
150077.pdf
Download (360kB)
Indexing & Metrics