针对雷达图像道路与水体难以区分的问题和多光谱图像易受干扰的现象，提出了一种基于图像融合的道路提取算法。在对图像进行降低噪声和几何配准等预处理后，首先从Landsat TM第1、2、3波段合成图像中选择样本作为支持向量机分类器的输入进行训练，在结构风险最小化原则下，实现Landsat TM图像的分类；然后利用SAR图像对起伏地物散射特征明显的优势，通过阈值分割处理，检测出待确定的区域；再利用设计的特征融合算子对同一区域的Landsat TM图像和ERS-2 SAR雷达图像进行融合处理；最后根据道路的影像特征进行特征的判别分类。实验结果表明，该算法较好地融合了两者的优势，实现了道路信息的有效提取。

Considering that it is difficult to distinguish the rivers and roads in synthetic aperture radar and the multi-spectral image is apt to be interfered, we proposed an image fusion based road extraction algorithm.After the noise was suppressed and the images were registered, we firstly chose some samples in Landsat TM image whose bands were 1, 2 and 3 as the input of the support vector machine classification for realizing image classification under the principle of minimizing risk.Then, we used threshold segment for ERS-2 SAR image to detect regions of interest, and fused the images from Landsat TM and ERS-2 SAR of the same area by using the designed fusion operator.At last, the road points were recognized according to the characteristics.The experiment results show that the proposed algorithm can integrate the advantages of the two images and extract the information of roads efficiently.