基于遗传优化谱聚类的图形分割方法

Image Segmentation Algorithm of Spectral Clustering Optimized by Genetic

传统的谱聚类方法使用k-means达到最后的聚类目的。k-means对初始条件敏感,易陷入局部最优,从而导致传统的谱聚类方法应用到图像分割时效果不太理想。将遗传算法用于优化谱方法的聚类阶段,提出一种以遗传算法优化普聚类的图像分割方法(Image Segmentation Algorithm of Spectral Clustering Optimization Based on Genetic,ISCOG)。在合成图像与真实图像上的实验表明ISCOG算法极大地提高了谱聚类算法的稳定性和聚类质量,证明了ISCOG算法的优越性。

The traditional spectral clustering methods use k-means to achieve the final clustering. But k-means is sensi- tive to initial conditions and easily plunges into local optimum, which influence the effect of image segmentation with spectral clustering method. This paper proposed an image segmentation algorithm of spectral clustering optimized by ge- netic algorithm（ISCOG）, using the GA instead of k-means in spectral clustering algorithm The experiments on synthetic images and real images show that ISCOG algorithm greatly improves the stability and clustering quality of the spectral clustering algorithm.