论文部分内容阅读
层次聚类是一种常用的聚类方法,但传统的层次聚类面临着计算复杂度较大、抗噪音和例外点干扰能力较弱的问题.本文以可能性聚类方法为基础,首先提出软边界球分算法,可实现对数据集合理分裂.随后将这一策略与分裂式层次聚类过程相结合,构造一种基于软边界球分的分裂式层次聚类算法(SHPDHC).SHPDHC 具有较低的计算复杂度.与此同时,它能较好地发现自然数据类,确定出合理的聚类数目,并能自适应划分出例外数据点.理论分析与对人工数据集的聚类实验结果证明了上述几点.最后我们将 SHPDHC 应用于一类阴影图像的分割中,同样取得良好效果.
Hierarchical clustering is a common clustering method, but the traditional hierarchical clustering faces the problem of large computational complexity, weak anti-noise and exceptional point interference ability.Based on the possibility clustering method, this paper first proposes Soft boundary ball algorithm, which can make the data set reasonably split.Then this strategy is combined with the split-level hierarchical clustering process to construct a split-level hierarchical clustering algorithm (SHPDHC) based on the soft boundary sphere.SHPDHC has Low computational complexity.At the same time, it can better find natural data types, determine the reasonable number of clusters and adaptively divide the exception data points.Statistical analysis and clustering experiments on artificial data sets The result proves the above points.Finally, we apply SHPDHC to the segmentation of a kind of shadow image, and also achieve good results.