LUV色彩空间中多层次化结构Nystrm方法的自适应谱聚类算法

Adaptive spectral clustering algorithm based on Nystrm method with multi-level structure in LUV color space

提出一种在LUV空间中基于多层次化结构Nystrm方法的自适应谱聚类算法。首先引入LUV色彩空间,避免了RGB色彩空间中色彩辨别阈对分割的影响,在纹理、边缘区域取得了更好的分割效果;其次将谱聚类算法中基于多层次化结构的方法和基于Nystrm采样的方法结合起来,有效减少了运算时间、解决了数据量较大时计算过程中内存溢出的问题;最后在K均值聚类中通过对特征间隙(eigengap)的分析,自适应地选择K值的大小,解决了自动确定聚类数目的问题。将提出的方法在LUV色彩空间中和RGB色彩空间中分别进行图像分割实验,结果表明在LUV色彩空间中取得效果更加理想。同时也将提出的算法与基于Nystrm方法的谱聚类算法(spectral clustering-Nystrm,SC-N)进行比较。实验结果表明,该算法在数据运算量、运行时间和分割结果上都优于SC-N方法。

In this paper, we propose an adaptive spectral clustering algorithm based on the Nystrom method with multi- level structures in LUV color space. First, we introduce the LUV color space, which can effectively avoid the influence of barely noticeable differences on the segmentation results, achieving better result in texture and edge regions. Second, we combine the spectral clustering algorithm based on multi-level structure and the Nystrsm method. Our approach can reduce the operation time and solve the problem of memory overflow. Finally, in K-means, through the analysis of the eigengap to adaptive select the value of K, this approach can automatically determine the number of clusters. The proposed method is applied to image segmentation, respectively, in LUV color space and RGB color space. The experimental results show that in LUV color space we can obtain even better results. The data computation and operation time as well as the segmentation result of the proposed algorithm are superior, compared to the spectral clustering algorithm based on the Nystrom method （SC-N）.