[Paper Review] Color Image Segmentation using Adaptive Particle Swarm Optimization and Fuzzy C-means

Narayana Reddy A, Ranjita Das|arXiv (Cornell University)|Apr 18, 2020

Remote-Sensing Image Classification3 references5 citations

TL;DR

This paper proposes APSOF, a novel color image segmentation algorithm that combines Adaptive Particle Swarm Optimization (APSO) with Fuzzy C-means (FCM) clustering to overcome FCM's sensitivity to initial conditions and noise. APSO optimizes cluster initialization by dynamically adjusting inertia weight and learning factors based on swarm behavior, significantly improving segmentation accuracy and convergence over standard FCM and PSO.

ABSTRACT

Segmentation partitions an image into different regions containing pixels with similar attributes. A standard non-contextual variant of Fuzzy C-means clustering algorithm (FCM), considering its simplicity is generally used in image segmentation. Using FCM has its disadvantages like it is dependent on the initial guess of the number of clusters and highly sensitive to noise. Satisfactory visual segments cannot be obtained using FCM. Particle Swarm Optimization (PSO) belongs to the class of evolutionary algorithms and has good convergence speed and fewer parameters compared to Genetic Algorithms (GAs). An optimized version of PSO can be combined with FCM to act as a proper initializer for the algorithm thereby reducing its sensitivity to initial guess. A hybrid PSO algorithm named Adaptive Particle Swarm Optimization (APSO) which improves in the calculation of various hyper parameters like inertia weight, learning factors over standard PSO, using insights from swarm behaviour, leading to improvement in cluster quality can be used. This paper presents a new image segmentation algorithm called Adaptive Particle Swarm Optimization and Fuzzy C-means Clustering Algorithm (APSOF), which is based on Adaptive Particle Swarm Optimization (APSO) and Fuzzy C-means clustering. Experimental results show that APSOF algorithm has edge over FCM in correctly identifying the optimum cluster centers, there by leading to accurate classification of the image pixels. Hence, APSOF algorithm has superior performance in comparison with classic Particle Swarm Optimization (PSO) and Fuzzy C-means clustering algorithm (FCM) for image segmentation.

Motivation & Objective

• Address the limitations of standard Fuzzy C-means (FCM) in image segmentation, particularly its sensitivity to initial cluster centers and noise.
• Overcome the drawbacks of traditional PSO by introducing adaptive parameter tuning to enhance convergence and cluster quality.
• Develop a hybrid framework that leverages APSO as a superior initializer for FCM to achieve more accurate and robust segmentation.
• Improve the reliability and performance of image segmentation in complex color images where standard FCM fails to produce satisfactory results.

Proposed method

• Apply Adaptive Particle Swarm Optimization (APSO) to dynamically adjust key hyperparameters—particularly inertia weight and cognitive/social learning factors—based on swarm behavior during optimization.
• Use the optimized particle positions from APSO as the initial cluster centers for the Fuzzy C-means (FCM) clustering algorithm.
• Integrate FCM with the improved initialization to refine pixel classification based on fuzzy membership and minimize an objective function measuring intra-cluster variance.
• Leverage the fast convergence and low parameter count of PSO while enhancing its performance through adaptive control mechanisms.
• Formulate the objective function for FCM as the minimization of weighted sum of squared distances between pixels and cluster centers, with fuzzy membership degrees.
• Iteratively update particle positions in APSO using velocity and position update rules informed by personal and global best positions, with adaptive parameter adjustment.

Experimental results

Research questions

• RQ1 Can adaptive parameter tuning in PSO significantly improve the initialization quality for FCM in color image segmentation?
• RQ2 Does the proposed APSOF framework outperform standard FCM and PSO-FCM in terms of segmentation accuracy and robustness to noise?
• RQ3 How does the dynamic adjustment of inertia weight and learning factors in APSO affect convergence speed and cluster quality?
• RQ4 To what extent does APSOF reduce dependency on initial cluster count and initial center selection compared to conventional FCM?
• RQ5 Can the hybrid APSOF approach produce visually superior and more accurate segmentation results on real-world color images?

Key findings

• The APSOF algorithm achieves superior segmentation accuracy compared to standard FCM and PSO-FCM due to improved initialization via adaptive parameter control.
• APSO reduces FCM’s sensitivity to initial cluster center selection by effectively exploring the search space and converging toward optimal configurations.
• The adaptive adjustment of inertia weight and learning factors enhances convergence speed and prevents premature convergence in the optimization process.
• Experimental results demonstrate that APSOF produces more accurate and stable cluster centers, leading to better pixel classification in color images.
• The proposed method outperforms both FCM and PSO-FCM in correctly identifying the true number of image regions and minimizing segmentation error.
• Visual evaluation confirms that APSOF yields more coherent and meaningful segmented regions than FCM, especially in noisy or complex color images.