1993 Volume 29 Issue 1 Pages 26-33
Remotely sensed images have different geometrical distortion caused by change of orbits or posture of platforms. Precise registration of these images is very important in multi-temporal processing. There have been reported a lot of registration methods for multi-temporal processing of images. These methods require ground control point (GCP) pairs for describing the relative geometrical discrepancy between images, and for determining the transform function from one to another. Selection of GCP pairs is essential in these methods. It is, however, difficult and time consuming to select them precisely. Inadequate selection of them causes registration error.
For reduction of error in registration and time for selecting GCP pairs, we propose an automatic registration method ART. In this method, we assume that only small change occurs in the scene during the period between the first and the second sensing. The proposed method ART consists of three stages of processing, that is, taking initial GCP pairs, detection and division, and local Affine transformation (ATF: Affine TransFormation).
In the first stage, we take three GCP pairs on both images for making initial triangles T1 and T2. In the second stage, residual discrepancy after ATF of T2 to T1 is evaluated in the detection procedure. We use spatial correlation as a key for measuring discrepancy in the sense of spatial structure. If the residual discrepancy is detected, T1 and T2 are divided into two sub-triangles. These procedures are iteratively applied to the triangles until the geometrical distortion between the images is sufficiently corrected by local ATF. The third stage of ART is application of local ATF to each divided triangle.
In this paper, we describe the principle and the procedures of the proposed method. The validity of this method is confirmed by numerical simulation and application to actual airborne images.
