随着遥感技术的发展，高分辨率遥感影像已经广泛应用于林业。然而遥感卫星在成像时，遥感卫星影像会受到多种因素的影响而发生几何畸变，因此必须对影像进行几何校正实现精确的地理编码，为后续的几何测量、相互比较及复合分析提供支持。在传统的摄影测量领域，应用较多的是物理模型。这种模型的处理技术已趋向成熟，定位精度比较高。但由于物理传感器模型涉及传感器物理结构、成像方式及各种成像参数。普通用户并不能得到这些信息，因此只能使用通用几何校正模型对遥感影像进行处理。 本文针对一景深圳地区QuickBird影像，分别论述了高分辨率遥感影像的各种通用几何校正模型及其解算方法，并就影响林区高分辨遥感影像几何精校正精度的因素做了相关研究，在此基础上分析了林区高分辨率遥感影像的变形规律。主要结论为： (1) 在所有的通用几何精校正模型中，有理函数模型的校正精度是最高的，但是有理函数模型需要的控制点数量比较多，而且模型的解算精度依赖于控制点的分布情况。 (2) 一般多项式模型的精度受地形起伏的影响较大，改进多项式模型的精度随多项式的阶数变化而变化,地形起伏的影响很小,选择合适阶数的改进多项式模型可以获得较高的几何校正精度。 (3) 在三种直接线性变换模型中，自检校直接线性变换模型的检查点点位中误差最小，而且利用少量的地面控制点就能得到稳定的解算结果。 (4) 在选择某一种方法进行林区高分辨率遥感影像几何校正的时候,应该综合考虑精度、算法复杂性、对已知数据的要求等多种因素。 (5) 对于林区高空间分辨率遥感影像的几何精校正,改进多项式模型精度较高、计算量较小、对控制点的数量和空间分布要求低,是一种较好的近似几何校正算法。 (6) 林区高分辨遥感影像变形受地形影响较大，几何校正时应充分考虑地形因素。

With the development of remote sensing technology, high-resolution remote sensing image have been widely used in forestry. However, the remote sensing image are affected by different errors and bring the geometric distortion. So the geometric rectification must be used in order to attain precise geocoding and provide the support for geometrical measuring，reciprocal comparison，image compound and etc. In conventional filed-photography, physical model that reaches maturity and has the feature of high resolution is mostly used. physical model is of limitation in application because of its sophisticated imaging eometry, complicate sensor physical structure，and self-correlation of model parameters．The universal geometry model is the only way to rectify the high-resolution remote sensing images because of lacking parameter of physical model. A scene of QuickBird image in Shenzhen was used in this paper, a variety of universal geometric rectification models and the solution method were discussed. The factors affect the geometric rectification accuracy of high resolution remote sensing image and the image distortion in forest areas were studied. The main conclusions are: (1) In all of the common geometric correction models, the rational function model of the calibration accuracy is the highest, but the rational function model needs more ground control points, and the accuracy of the model depends on the distribution of control points. (2) The precision of average polynomial method is worse for uneven terrain, varying greatly with different kinds of terrains, and the precision of improved polynomial methods varies with the order of polynomials and is nearly irrelative to the types of terrains. (3) Among the three direct linear transformation models, the precision of SDLT is the best， and it only need a small number GCPs. (4) The balance among precision, complexity, requirements for known data should also be considered for choosing methods from these approximate image rectification algorithms. (5) Experimental results also show that the improved polynomial method is a better choice for approximate rectification of high-resolution remote sensing image in forest area, from the viewpoints of precision, complexity, the number and spatial distribution of control - points and so on. (6) The distortion of high resolution remote sensing image in forest area depends on the terrain, so the factor of terrain must be included in the process of geometric rectification.