针对谱聚类集成算法计算复杂度高,难以应用到大规模图像分割处理的问题,提出一种将MS和基于超边相似度矩阵的谱聚类集成算法(HSMCESA)相结合的彩色图像分割算法(MS—HSMCESA)。首先,采用MS算法对彩色图像进行预分割,计算分割得到的每个区域的所有像素的彩色向量的平均值,以此作为HSMCESA的输入。在HSMCESA的谱分解过程中,通过矩阵变换对特征值分解进行近似求解,大大降低了算法的时间复杂度。对比实验表明:MS—HSMCESA较MS—Kmeans和MS-Ncut算法能获得更好的分割质量。
Aiming at problem that spectral cluster ensemble algorithm is hard to be applied in large scale image segmentation processing because of high computational complexity, a new color image segmentation method combining mean shift (MS) and Hyperedges' similarity matrix-based custer ensemble spectral algorithm (HSMCESA) named MS-HSMCESA is proposed. First, some regions are obtained through pre-segmentation by MS algorithm. The average value of color vectors in each region are considered as input of HSMCESA. Through matrix transformation, it computes eigenvalues of a small matrix to obtain the eigenvalues of the similarity matrix to reduce the time complexity. Experimental results show that MS-HSMCESA can always obtained better image segmentation quality than MS-Kmeans and MS-Ncut algorithm.