Image segmentation based on a new self-adaptive ant clustering algorithm

Image segmentation can be considered as the process of clustering image pixels of different image features. Clustering algorithm based on ant behaviors is a parallel, self-organized algorithm with sound discreteness, positive feedback and robustness. The basic ant colony algorithm is redundant in futile program loop, random in search mechanism and of which, the result is sensitive to the initial parameters. This paper improves these defects and suggests the idea of hierarchical clustering method. Then it applies the improved algorithm to image segmentation with the feature vector comprising of grayscale, gradient and neighborhood. Analytically, the improved algorithm has such merits as fast convergence, clustering effect of high quality, high clustering efficiency and robustness.