Abstract: In the image segmentation algorithm based on edge features, it is the premise and key to accurately detect the edge. Aiming at the shortcomings of remote sensing image segmentation algorithms, such as poor robustness, easy edge information loss and narrow application range, this paper proposes a remote sensing image segmentation algorithm based on improved Canny edge detection. For the traditional Canny Gaussian filter, the image is smoothed while smoothing the image, and the image is smoothed by the guidance filtering with the edge-preserving property. For the problem that the noise is sensitive to the derivative, the gradient template is added to calculate the image gradient and direction. For the limitation of setting the high and low thresholds by the traditional Canny operator, the Ostu algorithm is used to adaptively select the high and low thresholds according to the gray level of the image. In order to make full use of the advantages of multi-spectral images, band decomposition is performed in the edge detection, and processing is performed on a band-by-band basis, and then the edges are integrated into a result image, and finally region growth is performed. The experimental results show that compared with the traditional detection edge segmentation method, the proposed method achieves good results in remote sensing image segmentation.

Key words: image segmentation, edge detection, guided filter, Canny

摘要： 基于边缘特征的图像分割算法中，能够准确地检测出边缘是进行图像分割的前提和关键。针对目前遥感图像分割算法普遍存在鲁棒性差、易发生边缘信息缺失以及适用范围较窄的缺点，提出了一种基于改进Canny边缘检测的遥感影像分割算法。针对传统Canny高斯滤波在平滑图像的同时也模糊了边缘，改用具有保边特性的引导滤波对图像进行平滑；针对噪声对求导敏感这一问题，增加45°和135°方向梯度模板来计算图像梯度和方向；针对传统Canny算子人为设定高、低阈值的局限性问题，改用大律法自适应地根据图像灰度选取高、低阈值。为充分利用多光谱图像的优点，在进行边缘检测时采用波段分解，逐波段进行处理，随后将边缘综合成一幅结果图，最后进行区域生长。实验结果表明，与其他和传统边缘检测的分割方法相比，该方法在遥感影像分割中取得了较好的结果。

关键词: 图像分割, 边缘检测, 引导滤波, Canny算子