Research On Fast Registration Of The Remote Sensing Images Based On CUDA Parallel Computing

Remote sensing image registration is a key step to take full advantage of multi-source remote sensing image information and improve the effect of remote sensing application.With the requirement of remote sensing image registration precision becomes higher and higher,the complexity of registration algorithm is also constantly increasing and can't be applied to the situation with high real-time demand.Since CUDA has properties of low cost,low power,good portability,flexibility and so on,it is important and significant to research the CUDA parallel computing for remote sensing registration to improve the processing speed.This paper mainly studies the CUDA parallel implementation for some key steps of remote sensing image registration algorithm.The main works are as follows:(1)Refers to the C++ source code of SIFT feature extraction based on Open CV library,the functions of the Open CV library function are rewritten and realized to be independent of Open CV Library.(2)By analyzing the computation complexity of the SIFT feature extraction each steps,the accelerated SIFT feature extraction process of CUDA is designed,and others CUDA parallel implementation are detailed,such as the establishment of the gaussian pyramid,key point extraction,feature description and so on.The performance of the designed parallel implementation are verified by experiments.(3)The registration algorithm based on the combination of SIFT algorithm and other algorithms are more and more widely studied in multi-source remote sensing image registration.This paper mainly research on the parallel implementation of the image registration algorithm based on the SIFT and the optimization of regional mutual information.In addition to the SIFT feature extraction,the most time-consuming part of this kind of algorithm is the computation of the objective function in the iterative process.The crucial steps are the calculation of image interpolation matching and regional mutual information.Based on CUDA,the parallel implementation of region mutual information algorithm and image re-sample algorithm are introduced in detail,and the speed ratio of the algorithm is analyzed by experiments.