High level tasks in image analysis and understanding are based on accurate image segmentation which can be accomplished through multilevel thresholding. In this paper, we propose a new method that aims to determine the number of thresholds as well as their values to achieve multilevel thresholding. The method is adaptive as the number of thresholds is not required as a prior knowledge but determined depending on the used image. The main feature of the method is that it combines the fast convergence of Particle Swarm Optimization (PSO) with the jumping property of simulated annealing to escape from local optima to perform a search in a space the dimensions of which represent the number of thresholds and their values. Only the maximum number of thresholds should be provided and the adopted encoding encompasses a continuous part and a discrete part that are updated through continuous and binary PSO equations. Experiments and comparative results with other multilevel thresholding methods using a number of synthetic and real test images show the efficiency of the proposed method.

Image segmentation can be cast as a clustering task where the image is partitioned into clusters. Pixels within the same cluster are as homogenous as possible whereas pixels belonging to different clusters are not similar in terms of an appropriate similarity measure. Several clustering methods have been proposed for image segmentation purpose among which the Fuzzy C-Means clustering algorithm. However this algorithm still suffers from some drawbacks, such as local optima and sensitivity to initialization. Artificial Bees Colony algorithm is a recent population-based optimization method which has been successfully used in many complex problems. In this paper, we propose a new fuzzy clustering algorithm based on a modified Artificial Bees Colony algorithm, in which a new mutation strategy inspired from the Differential Evolution is introduced in order to improve the exploitation process. Experimental results show that our proposed approach improves the performance of the basic fuzzy C-Means clustering algorithm and outperforms other population based optimization methods.