医学图像分割中基于数据浓缩的谱聚类算法

摘要

基于传统Parzen窗密度估计函数的均值漂移谱聚类算法的时间复杂度不低于O(N2),不适合医学图像分割的实际需求.为此,通过压缩集密度估计和吸引盆均匀抽样两重数据浓缩策略以降低原MSSC的高时间开销问题,从而提出新的基于数据浓缩的谱聚类算法.实验结果表明,该算法能有效降低时间开销,较好地适应医学图像分割的要求.%The time complexity of the Parzen Window(PW) based Mean Shift Spectral Clustering(MSSC) algorithm is not less than OQf), which means that it is impractical for medical image segmentation. In is paper, the problem of heavy time cost of original MSSC is solved by using two strategies of data condensation: reduced set density estimator and random sampling from every attraction basin, and the novel Data Condensation Based Spectral Clustering(DCBSC) algorithm is proposed. Compared with MSSC, the time cost of DCBSC is decreased effectively, and the practicability of DCBSC for medical image segmentation is improved accordingly.