改进相似度的仿射传播聚类算法

Affinity Propagation Clustering Based on New Similarity

对舰船三维模型进行视点空间均匀投影,投影图像存在信息冗余,聚类技术是消除冗余投影的方法之一.由于缺少舰船投影的聚类知识,为了避免聚类结果受限于初始类代表点选择的缺点,研究了仿射传播聚类算法,首先提取所有投影图像特征,然后将所有特征进行归一化处理并作为初始聚类中心,提出了用空间向量模型计算特征相似度的方法,合并相似特征对应的投影图像,最后用聚类中心特征表示舰船目标.为了进一步验证改进相似度聚类算法的聚类质量,进行了聚类有效性分析,实验表明改进算法聚类质量好于原算法.

The projection images have redundancy information with even sampling ship 3D model in the view point space.Clustering technique is one of the methods for removing of redundant projections.Affinity propagation clustering can be used to divide all the views into a series of aspect ones,and overcome lacking of clustering knowledge and the weakness of selection initial parameters.Firstly extracting features from projections,then normalizing all the features and taking all the views as initial clustering centers.The paper introduced space vector model to compute similarity between views,and combined similar characteristics.At last obtained cluster centers for a ship.Assembling clustering validity index analysis,the improved algorithm is better than the original one in the clustering quality with experimental data.