图像分割的加权稀疏子空间聚类方法

Weighted-sparse subspace clustering method for image segmentation

在稀疏子空间聚类算法的基础上,提出一种基于加权稀疏子空间聚类的图像分割方法。利用加权的稀疏约束使得特征数据能够更好地被同一子空间内相似性高的特征数据线性表示,系数矩阵在类间更为稀疏。实验表明,给出的加权稀疏子空间聚类方法对于干净数据和带噪声的数据都能得到较高的数据聚类准确率,对自然图像能够得到比较符合人眼视觉特性的分割结果。

On the basis of sparse subspace clustering algorithm, a novel image segmentation method based on weighted-sparse subspace clustering is presented. By the constraints of weighted-sparsity, each feature data can be linearly represented by a few most similar feature data within the same subspace, and the resulting coeffi cient matrix sparse inter-class. Experiments show that the proposed weighted-sparse subspace clustering method can obtain higher clustering accuracy than the state of art methods for both clean and noisy data. Segmentation results by using this method on natural color images show good visual consistency.