Multi-Objective Image Segmentation Using a DEGA–PSO-SA Hybrid Metaheuristic

Image segmentation constitutes a foundational component of visual communication and image representation, where accurate structural delineation directly affects perceptual fidelity and semantic interpretability. Despite extensive advances in optimization-based segmentation, achieving a principled balance between global search capability, local convergence, computational efficiency, and representation stability remains an open challenge. This study introduces two hybrid metaheuristic frameworks—PSOSA (Particle Swarm Optimization integrated with Simulated Annealing) and DEGA (Differential Evolution combined with Genetic Algorithm)—designed to unify complementary exploration–exploitation dynamics within a coherent optimization-driven segmentation paradigm. By integrating swarm intelligence, evolutionary recombination, and probabilistic annealing mechanisms, the proposed methods enhance boundary discrimination, region homogeneity, and structural consistency in the resulting visual representations. Comprehensive evaluation on the Leeds Butterfly Dataset demonstrates that PSOSA achieves consistently superior Dice, Jaccard, and Structural Similarity Index (SSIM) scores, indicating improved preservation of perceptual structure and inter-region contrast. Conversely, DEGA attains competitive segmentation fidelity with substantially reduced computational cost, underscoring a quantifiable accuracy–efficiency trade-off. Statistical robustness analysis across independent trials further confirms enhanced stability and reproducibility, particularly for PSOSA. Collectively, the results establish hybrid metaheuristic optimization as a scalable and representation-aware strategy for image segmentation, providing a flexible decision framework adaptable to performance-critical and resource-constrained visual computing applications.