An Image Segmentation Algorithm Based on an Improved Particle Swarm Optimization

Traditional multi-level thresholod image segmentation methods, such as Otsu, consumes too high computational times, which are inadequate for dynamic scenes. As a population-based optimization algorithm, PSO could save the computational time of the segmentation methods. But the comparative weak global search ability of PSO still constraints its development. This study enhances the performance of PSO with quantum behavior through three key advancements: (1) Aiming to integer value operation for image segmentation, we adopt a random discrete search strategy to enhance the precise search; (2) A new updating strategy of individual with the best performance is proposed; (3) For better balancing the global and local search abilities, a Cauchy mutation strategy is presented. Tests conducted on the Berkeley Dataset demonstrate that this enhanced method achieves $7.2-12.8 {\%}$ higher fitness and 19.3-25.6% lower deviations compared to QPSO, outperforming the other compared algorithms.