Abstract: Among current clustering algorithms in remote sensing image processing, most of them have some unavoidable disadvantages, such as, difficulty to specify the number of classes, slow clustering performance, lack of stability and salt noised results. Based on object oriented image processing technique and the principle of competitive cooperative learning, this paper proposes an automatic image clustering algorithm, which does not need to specify the extract number of classes. In order to speed up the algorithm and obtain stable results, this paper proposes a novel dynamic median cut method to select the optimized initial cluster centers. Experiments show that the proposed clustering algorithm can obtain satisfying results and is valuable for application of automatic remote sensing image classification.

Key words: competitive cooperative learning, image clustering, image classification, remote sensing image processing, object oriented

摘要： 目前的遥感图像聚类方法通常存在一些不可避免的缺陷，如类别数难于自动确定、聚类速度缓慢、聚类过程不稳定以及聚类结果存在椒盐噪声等。结合竞争合作学习和面向对象的图像处理技术的优点，提出一种无需事先指定确切类别数的面向对象的竞争合作学习图像聚类算法。为了加快聚类速度并获得稳定的聚类结果，还提出一种基于动态包围空间的中位切分算法，用于选定初始聚类中心。通过对遥感影像的聚类实验，验证了该算法能够自动获得聚类数并得到满意的聚类结果，说明算法具有很好的实用价值。

关键词: 竞争合作学习, 图像聚类, 图像分类, 遥感影像处理, 面向对象