联合模型初始化独立谱聚类算法

Initialization-independent spectral clustering on the joint model

针对原始谱聚类算法初始化敏感的缺点,提出了一种基于联合模型的初始化独立谱聚类算法并将其用于图像分割.通过引入联合模型可以充分利用待聚类数据所包含的空间邻近信息和特征相似性信息,得到更精确的聚类结果;通过引入K-调和平均算法克服了原始谱聚类算法对初始化的敏感性,从而得到更稳定的聚类性能.最后,通过对纹理图像和合成孔径雷达图像分割验证了新算法的有效性.

Due to the initialization-dependence of original spectral clustering, an initialization-independent spectral clustering on the joint model is proposed and then is applied to image segmentation. The joint model can make full use of the information, spatial adjacency information and feature similarity information included in the data and then a more precise clustering result can be obtained. And the introduction of the K-Harmonic Means algorithm （KHM） can overcome the initialization-dependence of original spectral clustering and thus a more robust clustering result can be obtained. Experiments on textural images and Synthetic Aperture Radar （SAR） images verify the validity of the proposed algorithm.