Self-adaptive Multiple Evolution Algorithms for Image Segmentation Using Multilevel Thresholding

Multilevel thresholding based on Otsu method is one of the most popular image segmentation techniques. However, when the number of thresholds increases, the consumption of CPU time grows exponentially. Although the evolution algorithms are helpful to solve this problem, for the high-dimensional problems, the Otsu methods based on the classical evolution algorithms may get trapped into local optimal or be instability due to the inefficiency of local search. To overcome such drawback, this paper employs the self-adaptive multiple evolution algorithms (MEAs), which automatically protrudes the core position of the excellent algorithm among the selected algorithms. The tests against 10 benchmark functions demonstrate that this multi-algorithms is fit for most problems. Then, this optimizer is applied to image multilevel segmentation problems. Experimental results on a variety of images provided by the Berkeley Segmentation Database show that the proposed algorithm can accurately and stably solve this kind of problems.