Convergent heterogeneous particle swarm optimisation algorithm for multilevel image thresholding segmentation

One of the critical tasks in image processing is image segmentation. Image thresholding is the simplest technique of segmentation in two forms of bi-level and multilevel. One alternative to find optimal threshold values is to convert the problem of segmentation into an optimisation problem. Classical optimisation techniques are computationally expensive, inaccurate and inefficient compared to the recent global heuristic optimisation algorithms. In this study, Convergence heterogeneous particle swarm optimisation (PSO) algorithm, has been utilised to find the optimal multilevel thresholds. The general idea of this algorithm is to divide particles into four subswarms for searching problem space. Otsu's and Kapur's thresholding methods are separately used as a fitness function which the former maximise between-class variance and the latter maximise image entropy. To evaluate the proposed method, it applied to a benchmark of images and the results compared with similar and famous heuristic methods, genetic algorithm, harmony search and the PSO. The results revealed that the proposed method is accurate and robust whereas through several executions, it shows more stability with better convergence in compare to the other approaches while difference was significant by increasing the number of thresholds.