Research On High Resolution Multispectral Remote Sensing Image Registration On Board Unmanned Airship

Low-cost high resolution multi-spectral image is attractive to customers in many fields. This thesis introduces a unmanned airship high resolution multi-spectral remote sensing system, which is curruntly being developed, and our primarily study is the automatic registration of multi-spectral image.Firstly, this article introduced the its general frame, main function and devices of the multi-spectral imaging system .Secondly, based on the study on the principle and characteristic of two-dimensional fourier transform, we optimize the process of Fourier-Mellin transform image automatic registration method. When using image fourier frequency spectrum to calculate rotation angle and transformation scale factor,it is always needed to make Fourier-Mellin transformation on amplitude spectrum. This mathod is different from traditional uniform sampling or Pseudo-polar Coordinate Transformation. The adaptive resampling is introduced in the process of polar coordination, which controls the number of sampling points based on changing polar radius length. This method can effectively overcome over-sampling in low frequency region, improve accuracy and effect of the multi-spectral image registration, and it has well solved the problem of the multi-spectral image registration under simple image motion model, and meanwhile realize the the multi-spectral image registration in low accuracy condition.At last, in order to improve the multi-spectral image registration accuracy, especially aiming at resolving image partial distortion under perspective imaging condition, after analysing the properties of classic homography, combined with the characteristics of unmanned airship high resolution remote sensing system introduced in this thesis, we put forward a method of image registration on the theory of multi-homography matrix model. Comparing with the traditoinal Two-Image method solving homography matrix factor, the multi-image simultaneous solving method, which solves homography matrix factors simultaneous, is an global image registration method which can fully reflect the advantage of high accuracy of the least square method in data processing. In addition, introducing Fourier-Mellin transformation into the process of the extraction and matching of image homonymy characteristics greatly increased the effectivity and reliability of characteristics points searching and matching, thus on the whole improved the accuracy and effective of the multi-spectral image registration.