关键词: 谱聚类, 均值漂移, 极化SAR, 图像分割, 边缘检测

Abstract:

Image segmentation by spectral clustering is a clustering method based on points. It is characterized by the use of similarity measure matrixes. We usually need to calculate the similarity matrixes between every two cells, which consumes huge computation task and a lot of time when processing large images. To solve this problem, we propose an improved method. First, we use mean shift algorithm for polarimetric SAR image, and then select center pixel to construct similarity measure matrix. At last, we use the normalized segmentation rule for image segmentation. Computation experiment proves that the algorithm could improve the efficiency with high accuracy and satisfactory result, hence is of practical significance.

Key words: spectral clustering, mean shift, Polarimentric SAR, image segmentation, edge detection