【作者】 张秀春； 蔡敏； 张运杰；

【Author】 Zhang Xiuchun;Cai Min;Zhang Yunjie;School of Science,Dalian Jiaotong University;School of Science,Dalian Maritime University;

【机构】 大连交通大学理学院； 大连海事大学理学院；

【摘要】 本文针对近邻传播(affinity propagation,AP)聚类算法在构造相似度矩阵时未考虑图像内蕴几何量的不足,提出采用测地距离算法构造相似度矩阵。首先,在尽可能地考虑图像与图像行列之间的结构信息的基础上求解出两点间的曲面距离,使得图像信息更加完善;其次,针对AP算法运算复杂度较高的缺陷,运用Nystrm逼近策略降低求解相似度矩阵的运算复杂度,降低内存消耗;最后,仿真实验表明无论是从客观角度还是主观角度,本文算法均获得比传统AP聚类算法更好效果。更多还原

【Abstract】 Based on the defect of affinity propagation clustering algorithm which doesn’t take the pictures’ intrinsic geometry into consideration when doing the similarity measure,a new similarity measure of geodesic distance is proposed in this paper.First,try to get the distance of a curved surface between two points using the structural information between the ranks of pictures to make the pictures’ information more perfect;then based on the big defect of affinity propagation clustering algorithm,we use the Nystrom approximating strategy to reduce the algorithm complexity of getting Similarity Matrix and the memory consumption.As a result,it is proved in this experiment that the new algorithm in this paper could offer a better segmentation result than conventional affinity propagation clustering algorithm.更多还原