Manifold affinity propagation clustering for PolSAR image classification

A novel affinity propagation (AP) clustering method based on combined manifold distance (CMD-AP) is proposed for polarimetric SAR (PolSAR) image classification. It solves the problem that traditional AP clustering algorithm, which employs Euclidean distance as similarity measures, cannot effectively process the complex PolSAR data. After dividing the PolSAR image into superpixels, texture features as well as corresponding polarimetric features are extracted from PolSAR image, and Laplacian eigenmap (LE) is used to reduce the dimension of the features. Then Wishart manifold of coherency matrix and Euclidean manifold of feature vector are combined to form the manifold distance measures. Finally, the PolSAR image is classified by AP clustering algorithm. The proposed method considers the potential manifold structure of PolSAR data and introduces the combined manifold distance measures into AP clustering. Experimental results demonstrate that the proposed method improves the classification performance, has advantages in region homogeneity and edge preservation.