雷达图像的线条特征提取算法一般分为三步，首先作图像预处理，然后采用特定的边缘检测算子提取边缘点，第三步形成各种有意义的线条特征，并且将断裂的长线连起来。常见的提取算法都是基于光学图像的，所以在确定边缘点时，假定图像中的噪声是加性高斯白噪声。这样，在合成孔径雷达（SAR）图像中使用光学图像中提取边缘点的方法就不行了，这是因为SAR图像中的相干斑噪声是服从K分布的。不过，将边缘点连接为有意义的线条特征的方法还可以沿用。所以，我们可以采用两步检测算子来检测边缘点，然后使用从相位编组思想演化而来的方向编组法形成直线条特征方法。大量的实验验证了这种方法对于SAR图像是切实可行的。

The algorithm of extracting linear features from radar images generally has three steps. The first step is to pre-process images, the second step is to extract edge points using certain edge detectors, and the third step is to form all kinds of linear features and at the same time to connect splitting lines. Traditional methods of extracting linear features from optical images fail in processing SAR images. So this paper carries out edge extraction of SAR images because they often rely on the assumption that the noise is additive and white Gaussian, and the speckle noise in SAR images is K distributed. However, the linear feature extraction method based edge points can also be used. Therefore, two-step detector is used to detect edge points and then an orientation-grouping algorithm derived from the idea of gradient-based segmentation is applied to form linear features. We have obtained excellent effects through a great deal of experiments.

TN957.52