Treffer: Color image segmentation based on multi-level graph partitioning using ant colony optimization.

In order to eliminate the restraint of clustering number to segmentation result of spectral clustering based normalized cut image segmentation, this paper proposed an ant colony optimization based multi-level graph partition algorithm which was used to segment color natural landscape image. The proposed algorithm regarded similarity graph corresponding to image as ant colony's habitat, and then grouped similar vertices into partitions gradually under the guidance of normalized cut criterion using ant's foraging behavior, which completed graph partition problem by a multi-level way. To reduce calculated quantity of image segmentation, it utilized an effective super-pixel algorithm to preprocess original image. Contrast experiment shows that the proposed algorithm eliminates the restraint, while improves accuracy and speed of image segmentation based on normalized cut criterion. [ABSTRACT FROM AUTHOR]