改进的k'-means算法及其在图像分割中的应用

Improved k'-means algorithm and its application in image segmentation

k'-means是对k-means算法的一种改进,它引入了竞争惩罚学习机制,可以在无监督的情况下确定聚类数目.本文提出了两种新的基于频率敏感差异度量的k'-means算法,新算法利用竞争惩罚学习机制确定聚类数目.针对一组合成数据进行对比实验,结果表明新的k'-means算法可以成功地对数据集进行分类.最后,本文将新算法应用于图像分割.

k＇ -means is a improved k -means algorithm, it introduces competition punishment learning mechanism that can determine the unsupervised clustering number. This paper proposes two new k＇ -means with frequency sensitive discrepancy metrics. The new algorithms determine the number of clusters by competition punishment learning mechanism. Applied in synthetic datasets for a comparative experiment , the results show that the new k＇ -means can successfully classify the datasets. Finally, the new algorithms are applied to image segmentation.