Meteorological Automatic Image Registration Method

Image registration in remote sensing field means accurately matching two or more remote sensing images of the same region, which is an important issue in remote sensing image processing. As one of RS image registration, the research on meteorological image registration has significant importance in practical applications.This paper reports the research into meteorological image registration. The goal of this paper is to design some meteorological image registration methods with high automatization and the balance between precision and speed. Based on summarizing and digesting the researching findings by reasearchers from both home and abroad, several registration algorithms are explored thoroughly and are employed to solve the problems in meteorological image registration. To be specific, the researcher conducted studies from the following aspects:(1) Through studying correlation algorithm, a correlation matching algorithm is used to solve the automatic meteorological image registration. Firstly, Corner points are extracted by using SUSAN operator. Then corresponding points are found by using the correlation between two images with the help of relaxation algorithm to improve matching precision.(2) To meet the demand of fast and automatic meteorological image registration, an automatic meteorological image registration method is designed by using grid. When selecting RCP(Registration 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; a gradual matching algorithm is used for improving precision.(3) On the basis of wavelet multi-scale principle, an automatic meteorological image registration method is presented within the scope of wavelet transformation. First, control points are extracted by using wavelet transformation. Then wavelet transformed pyramids and cross-correlation algorithm are used for matching.Several experiments are carried out to register meteorological images with these methods. The result shows that these three methods have high precision, rapid computing efficiency and high automatization, which also indicates the validity and application value of the research result.The last part of this paper summarizes the work of the paper, analyzes the improvements needed to be done and gives the directions of future work.