Elastic Image Registration Method In Meteorological Images

Image registration is a hot topic in computer vision and pattern recognition, it has great values in medical imaging and multimodal image fusion. Image registration which can map points (or surface) from one image to homologous points on an objects in the second images is one of the fundamental tasks in image processing, and then fusion them into a new image.In this paper, in allusion to the imaging of the tiros which produce rigid deformation and nonrigid deformation, we review different methods in image registration, include rigid and nonrigid registration, and propose a few amelioration as followed.1. Due to the characteristic of the weather image, a hierarchical transformation model of the motion of the image has been introduced. The global motion of the image is modeled by an affine transformation which use the rigid registration based on Forstner interesting operator, while the local image motion is described by a free-form deformation (FFD) based on B-splines.2. In the rigid registration process, we improve on the distance restrict algorithm. A suitable threshold adjust method is add to the conventional algorithm. Compare with the conventional distance restrict algorithm, the improving method is more all-purpose, and can better removing the fake matching points.3. In the nonrigid registration process, we propose a new strategy to updating the control point grid (CPG) and the grid area intensity, which update the CPG and local area intensity step by step,and the result is using in the next computing. Experiments prove that the new strategy decrease the runtime of the nonrigid registration visible.In this paper, the algorithm has been applied to the fully automated registration of weather images, the result clearly indicate that the approach we proposed can well recover the motion and deformation of the weather images.