Image Clustering Method based on Particle Swarm Optimization

Abstract

To implement efficient computer vision mechanisms, efficient image clustering methods are important. The paper elaborates a clustering method based on particle swarm optimization (PSO) which provides automatic establishment of clustering parameters. The developed PSO based clustering method was tested on 860 images for a car vision system and its results and contribution to the pattern recognition quality improvement were assessed in comparison with fuzzy C-means and k-means. The results do not differ significantly, but distinction in average time of work for these methods was noted. The PSO clustering method is faster than k-means and slower than fuzzy C-means. However, fuzzy C-means method does not guarantee correct results during the further analysis, so the PSO clustering method can be more efficient for implementation in computer vision systems.