Study On High Performance Algorithm Of Image Registration For Remote Sensing

Image registration in remote sensing field means accurately matching two or more remote sensing images of the same ground region, which is an important issue in remote sensing image processing. Image registration is a computational intensive task. With the development of the technology in remote sensing, the volume of remotely sensed data increases tremendously. So the procedure of the registration becomes longer, and comes to a bottleneck task in the process of remote sensing images. Many experts have devoted themselves to the research of registration.Firstly, existing image registration algorithms are reviewed and their advantages and drawbacks are analyzed in this paper; then a high performance registration algorithm is studied. In the registration algorithm, a new method of LSSP (Local Search Strategy based on Pyramid) is proposed to carry on the search of the corresponding points, and a method named RSLE (Resampling based on the Local Edge) is adopted to implement the resampling for large remote sensing images. The characteristics of this algorithm are as follows:(1) When selecting RCP(Registation Control Point), coarse and fine grids are used to ensure that there are more RCP points in the regions with rich information than other regions, which makes the selected control points more effective for registration.(2) The LSSP adopts a new local search strategy, which reduces the search space, and saves searching time.(3) RSLE algorithm does resampling based on computing local output area. This algorithm is suitable for parallel processing and out-of-core computing..In the end, we implement the parallel registration algorithm. The experiment results show that our algorithm improves the registration accuracy efficiently, and also has good parallel performance.