An efficient image segmentation method based on expectation maximization and Salp swarm algorithm.

Area of Science:

• Medical Image Analysis
• Computational Intelligence
• Computer Vision

Background:

• Multilevel image thresholding is crucial for image segmentation.
• Expectation Maximization (EM) is efficient but suffers from local optima and inability to guarantee class number.
• Gaussian Mixture Models (GMM) are used for histogram estimation in EM.

Purpose of the Study:

• To develop a novel thresholding approach overcoming EM's limitations.
• To enhance medical image segmentation using a hybrid EM-SSA method.
• To ensure the desired number of clusters is maintained during segmentation.

Main Methods:

• A hybrid approach combining Expectation Maximization (EM) and Salp Swarm Algorithm (SSA) for multilevel image thresholding.
• Utilizing SSA to guide EM towards better solutions and escape local optima.
• Implementing a mechanism to maintain the specified number of clusters.

Main Results:

• The proposed EM-SSA method demonstrated an average improvement of 5.27% in Peak Signal-to-Noise Ratio (PSNR) and 2.01% in Feature Simulation (FSIM) compared to traditional EM.
• On CT scan images, the method achieved the top rank for PSNR and second rank for FSIM against four other segmentation techniques.
• Qualitative and quantitative results show superior segmentation performance compared to existing state-of-the-art methods.

Conclusions:

• The combined EM-SSA algorithm offers a robust and effective solution for multilevel image thresholding.
• This novel approach significantly enhances medical image segmentation accuracy and reliability.
• The method provides a promising alternative for complex image segmentation tasks in medical imaging.