In this work, three approaches (PSO, DE and GA) were proposed to find the suitable thresholds of an image, based on the concept of fuzzy c-partition and maximum entropy principle. The image can hold as much information as possible when it is transformed to the fuzzy domain. The algorithms are extended to multiple thresholding. The use of PSO, GA and DE algorithms reduce greatly the time complexity. The three approaches were then compared by testing them on various images. The algorithms are comparable in terms of solution quality for c=2. While the value of c increases, PSO, DE and GA provide the same results in terms of accuracy and robustness of the results, but in terms of execution time the GA is most efficient. The experimental results have shown the effectiveness and usefulness of the proposed algorithms for image segmentation. The PSO, DE and GA approaches can deliver satisfactory performance.