Journal of Frontiers of Computer Science and Technology ›› 2020, Vol. 14 ›› Issue (8): 1348-1357.DOI: 10.3778/j.issn.1673-9418.1909084

Previous Articles     Next Articles

Soft Subspace Clustering Algorithm Optimized by Brain Storm Algorithm for Breast MR Image

FAN Hong, SHI Xiaomin, YAO Ruoxia   

  1. School of Computer Science, Shaanxi Normal University, Xi??an 710119, China
  • Online:2020-08-01 Published:2020-08-07

头脑风暴算法优化的乳腺MR图像软子空间聚类算法

范虹史肖敏姚若侠   

  1. 陕西师范大学 计算机科学学院,西安 710119

Abstract:

The traditional soft subspace clustering algorithm is very susceptible to the initial clustering center and noise data when segmenting breast MR images with large amount of information, uneven intensity and boundary blur, which results in that algorithm falls into local optimum and causes serious misclassification. Aiming at solving this problem, a soft subspace clustering algorithm improved by brain storm algorithm for breast MR images cluster-ing is proposed in this paper. Firstly, a new objective function combines relaxation criterion and generalized noise clustering, and the membership degree calculation method is used to find the subspace where the cluster class is located. Then, the clustering task in the subspace is adapted with a given index. Finally, the brain storm algorithm is used in the clustering process to balance local search and global search and overcomes the disadvantages that the existing algorithms are easy to fall into local optimum. The experimental results of the comparison algorithms and the proposed algorithm in Berkeley image dataset show that the proposed algorithm has higher precision, and the clustering results of clinical breast MR images verify the strong robustness of the proposed algorithm.

Key words: breast MR image, brain storm algorithm, soft subspace clustering algorithm, image clustering

摘要:

传统的软子空间聚类算法在对信息量大、强度不均匀、边界模糊的乳腺MR图像进行分割时,易受初始聚类中心和噪声数据的影响,导致算法陷入局部最优,造成误分类。针对该问题,提出一种头脑风暴算法优化的乳腺MR图像软子空间聚类算法。算法首先引入一个放松界约束与广义噪声聚类结合的目标函数,并用隶属度计算方法来寻找簇类所在子空间;然后在子空间聚类时用给定指数来适配聚类任务;最后在聚类过程中运用头脑风暴算法进行优化,有效地平衡局部搜索与全局搜索,从而弥补现有算法易陷入局部最优的不足。对比算法与该算法在Berkeley图像数据集上的实验结果表明该算法具有较高的精度,临床乳腺MR图像聚类的实验结果验证了所提算法的鲁棒性。

关键词: 乳腺MR图像, 头脑风暴算法, 软子空间聚类算法, 图像聚类