Multi-Threshold Image Segmentation Based on the Hybrid Strategy Improved Dingo Optimization Algorithm.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Image Processing

Background:

• Standard optimization algorithms like the Dingo Optimization Algorithm (DOA) often struggle with local optima, slow convergence, and boundary handling in complex problems.
• Existing methods require enhanced mechanisms for global exploration and precise local exploitation to improve efficiency.

Purpose of the Study:

• To introduce the Hybrid Strategy Improved Dingo Optimization Algorithm (HSIDOA) to overcome the limitations of the standard DOA.
• To enhance convergence accuracy, speed, and stability in complex optimization tasks.

Main Methods:

• HSIDOA integrates quadratic interpolation search, horizontal crossover search, and centroid-based opposition learning for boundary handling.
• The algorithm was tested on CEC2017 (30-D) and CEC2022 (10/20-D) benchmark suites against seven mainstream algorithms.
• HSIDOA was applied to multi-level threshold image segmentation using Otsu's method as the objective function.

Main Results:

• HSIDOA demonstrated superior performance in average fitness, standard deviation, convergence rate, and Friedman rankings on benchmark suites.
• The algorithm exhibited strong robustness and scalability across various dimensions.
• In image segmentation, HSIDOA consistently achieved the best quality across multiple threshold levels (4-10), surpassing comparison algorithms in convergence speed and stability.

Conclusions:

• HSIDOA offers significant improvements in global exploration, local exploitation, convergence speed, and high-dimensional robustness.
• The algorithm is an efficient, stable, and versatile tool for complex numerical optimization and image segmentation.
• HSIDOA provides a robust framework for addressing challenging optimization problems in various domains.