The P2L method of mismatch detection for push broom high-resolution satellite images

RANSAC-based mismatch detection methods are widely used in the geometric registration of images. Despite their prevalence, setting the detection thresholds for different situations continues to be difficult without an appropriate geometric model. In high-resolution satellite images, simple image-space transformations are commonly influenced by the terrain or elevation errors. This paper introduces a new method, called the P2L method, which uses the distance between the transformed right image point and the segment of the corresponding epipolar line to distinguish the correct matches and mismatches. The affine model of the P2L method is solved to transform the right image points towards the segment of the epipolar line. The images for demonstration were acquired by GeoEye-1, Ikonos-2, and Ziyuan-3; and each type of image pairs had different intersection angles to explore the influence of the elevation error. The correct matches were manually collected and the mismatches were simulated. The experiments in this paper, which used only correct matches, demonstrated that this method was very robust with one specific threshold (five pixels) and was suitable for all the image pairs. The experiments using simulated mismatches and real matching points demonstrated that this method was able to distinguish most of the mismatches; and even for the image pair that had a 54-degree intersection angle, the ratio of mismatches was reduced from 81% to 11%.