Abstract: Affinity propagation is often limited by its inability to cluster datasets with inherent manifold structures. A novel clustering method, namely Affinity Propagation with Laplacian Eigenmaps(APPLE), is proposed to address this problem. It enhances the standard affinity propagation with manifold learning capacity. Geodesic distance is used to compute affinity between data points. Laplacian eigenmaps are applied to reduce the dimensionality and to extract features. Experimental results show APPLE outperforms standard affinity propagation in application of image clustering.

Key words: Laplacian eigenmaps, Affinity Propagation(AP), Dijkstra algorithm, Normalized Mutual Information(NMI)

摘要： 仿射传播方法难以处理具有流形结构的数据集。为此，提出一种基于拉普拉斯特征映射的仿射传播聚类算法(APPLE)，在标准仿射传播的基础上增强流形学习的能力。使用测地距离计算数据点间相似度，采用拉普拉斯特征映射对数据集进行降维及特征提取。对图像聚类应用的实验结果证明了APPLE的聚类效果优于标准仿射传播方法。

关键词: 拉普拉斯特征映射, 仿射传播, Dijkstra算法, 归一化互信息

CLC Number: 

• TP18