Comparison and analysis research on geometric correction of remote sensing images

The algorithms of remote image approximate geometric correction are mainly based on the least squares method (LSM) about linear or nonlinear models. Their disadvantages lie in overfitting, poor generalizing ability, and enough amount samples demand, due to the principle of the empirical risk minimization (ERM). It is put forward that the geometric correction algorithm of remote image making's use of support vector machine, combined with the essence theory of image approximate geometric correction. One testing region is selected; the coordinates of the ground control points in the remote image and in the ground are measured. Varying number control points are selected to correct the remote image. Other control points serve as testing points, by the cluster algorithm. The approximate geometric correction algorithm, neural network, and support vector machines algorithm are applied to geometrically correct the images respectively, and the comparison analysis of the correction accuracy is obtained.