A remote sensing image registration method combining feature information clustering and partitioning
Aiming at the problem that the global registration model can't correctly fit the local region due to topographic relief and rich ground object types,this paper proposes a method to quickly divide the image region and realize fine fitting based on feature information hierarchical clustering method.This method uses the scale constraint of difference space to extract the feature points of sift with higher accuracy,and optimizes the matching efficiency combined with Hellinger transform to complete the rough feature matching.The initial clustering is completed according to the point neighborhood information,and different models are obtained;the coincidence degree of matching points to different transformation models is calculated,the tendency set is constructed,the set is merged according to the distance to obtain the cluster center;and the sub region is generated using Tyson polygon method.The transformation model of each sub region is solved and interpolated to obtain the registration results.The remote sensing images of farmland,mountainous areas and coastal cities and towns are used for experiments.The registration effects of SIFT+ST,FSC-SIFT and PSO-SIFT methods are compared with this method.The results show that the accuracy and visual registration effect of this method are better.
国家科技期刊平台
NSTL
万方数据
维普
