Multilevel threshold image segmentation based on a novel mechanism enhanced coati optimization algorithm.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Meta-heuristic algorithms excel at multilevel threshold image segmentation.
• Existing methods struggle with balancing global/local search, premature convergence, and multi-task segmentation.

Purpose of the Study:

• To develop an enhanced optimizer for multilevel threshold image segmentation.
• To address limitations in handling multiple objective functions, diverse image types (grayscale/color), and high threshold levels.
• To improve global exploration and local exploitation balance and prevent premature convergence.

Main Methods:

• Proposed a novel search mechanism, ASSM (Salp Swarm Optimization Algorithm-inspired), for DP-algorithms.
• Developed the ENsemble Collaborative Optimizer (ENCOA) framework using Hierarchical Vertical-Horizontal Search (HVHS).
• Integrated improved circle chaotic mapping, opposition-based learning, and Lévy flight strategies into ENCOA.

Main Results:

• ENCOA demonstrated superior performance on CEC2017 benchmark and engineering problems.
• Applied to segment grayscale and color images using Kapur's entropy and Otsu's variance.
• Achieved higher convergence accuracy and segmentation quality, especially for high threshold levels (4-32).

Conclusions:

• ENCOA effectively overcomes limitations of existing meta-heuristic algorithms for image segmentation.
• The proposed framework offers significant improvements in accuracy and efficiency for complex segmentation tasks.
• ENCOA shows strong potential for advanced image segmentation applications.