Abstract: As basic content of images, the edge information has important value in the recognization of the typical ground targets. In order to reduce the speckle’s adverse influence on the edge abstraction of high resolution TerraSAR-X images, a combined method based on the mean shift and rough set theory is proposed. Firstly, the mean shift algorithm is carried on to restrain the speckles, and then the rough set theory based edge enhancement is applied to the edges detected by the sobel operator. Experiments with 1m TerraSAR-X images of man-made targets indicate that mean shift method does better than the traditional methods for the speckle reduction and edge preservation, meanwhile the enhancing process based on the rough set theory can efficiently improve the results of the sobel edge detectors, which in turn produces high quality edge abstraction for the typical ground targets.

Key words: TerraSAR-X image, speckle noise, mean shift, rough set theory, edge enhancement

摘要： 作为图像的基本内容之一，边缘信息在房屋等典型地面目标识别中具有非常重要的价值。针对相干成像机制所产生的斑点噪声在高分辨率TerraSAR-X边缘信息提取中的不利影响，给出了一种基于均值漂移和粗糙集理论的组合处理方法。首先以均值偏移显著抑制斑点噪声干扰，然后利用粗糙集理论的下近似和上近似关系，对Sobel边缘检测结果进行增强处理。房屋等人造目标的TerraSAR-X 1m分辨率图像的实验结果显示，均值漂移方法能够在有效抑制斑点噪声的同时，较好地保持边缘细节，而基于粗糙集的边缘增强可以显著改善Sobel边缘检测结果，最终生成高质量的目标边缘图。

关键词: TerraSAR-X图像, 斑点噪声, 均值漂移, 粗糙集理论, 边缘增强