Study On Automatic Registration Algorithm And Parallel Implementation For Hj-satellite Optical Imagery

HJ-satellite is one of satellites of"Environmental Monitoring and Disaster Forecasting satellite constellation",which is to dynamic monitoring the environment and disasters of our country, neighboring countries and regions. It requires fast and timely processing and application of these images. Image registration is the foundation of image accurate correction, mosaic, classification, thematic mapping, as well as study of remote sensing application. It plays a very important role in the field of image processing. To meet the need of image accurate correction, in this thesis, considering the characteristics of HJ-satellite images, this thesis studies the automatic, rapid, and accurate registration methods for HJ-satellite images. The main research contents are as follows:(a) HJ-satellite optical images registration techniques at gray-level. This thesis created the registration algorithm which is gray-level registration algorithm based on Improved CSS corner features detection for HJ-satellite optical images. Comparing the results of the created algorithm with gray-level registration algorithm based on Harris and CSS corner detection separately through experiments, this paper analyzed and summarized the advantages and limitations.(b) HJ-satellite optical images registration techniques at feature-level. This thesis developed the SIFT feature registration algorithm with C language, which can generate good results between HJ-satellite optical images matching by experiments and the accuracy can up to 1 pixel, and also suitable for images from different sensors and the accuracy can up to 2 pixels.(c) Design and realize parallelization for SIFT feature-level registration algorithm. To improve the efficiency of image registration, this thesis paralyzed the SIFT feature registration algorithm based on MPI parallel processing technology. First, analyze the parallelization of the SIFT feature registration algorithm. Second, the parallel program for this algorithm is developed. Last, the performance evaluation of the parallel program is given.