同大多数的线阵推扫式影像相比,TM影像作为双向扫描型的影像其几何校正方法有其特殊性和复杂性。针对这个问题,本文简要介绍了一般多项式、空间斜墨卡托投影、有理函数模型等几种几何近似校正算法,并利用广州地区的TM影像进行了各种试验分析和精度比较。结果表明:同其他方法相比,空间斜墨卡托投影是一种较好的算法,校正精度最高,能够达到一个像素左右,而且该方法不受地形起伏的影响,无需地面高程信息。 Compared with most of linear image, the TM image has its peculiarity and complexity as the method of geometric correction for bi-directional scanning image. To solve this problem, this paper briefly introduces several geometric approximation algorithms, such as general polynomials, space diagonal Cartesian projection, rational function model and so on, and makes a variety of experimental analysis and accuracy comparison using TM images in Guangzhou. The results show that compared with other methods, space diagonal Cato projection is a better algorithm with the highest correction accuracy and can reach about one pixel, and the method is not affected by the terrain fluctuation without the need of ground elevation information.