| The research of pan-sharpening algorithms is of great importance.With the help of those algorithms,we are able to merge the multi-spectral image with low spatial resolution and the panchromatic image with high spatial resolution,both shot by a satellite,generating a multi-spectral image with high spatial resolution.It can facilitate the accomplishment of remote sensing tasks like object detection.Existing popular pan-sharpening algorithms can be roughly classified as four categories: component substitution,multi-resolution analysis,variational model based methods,and neural network based methods.The main work in this thesis is building new pan-sharpening algorithms on the basis of those known.This is achieved according to two research lines.On the one hand,inspired by regularization techniques based on framelet and a variational model based pan-sharpening method utilizing reproducing kernel Hilbert space(RKHS),a new variational model is proposed.In the proposed model,assumption is made that the coefficients obtained after implementing framelet transform on the desired image are sparse,and the relationship between the multi-spectral image and the desired image,as well as the relationship between the panchromatic image and the desired image,is considered.The model can be solved efficiently by alternating direction method of multipliers(ADMM).In order to improve the performance of the algorithm,an outer iterative strategy is conbined with the proposed model.Numerical experiments indicate that this approach can gain many spatial details from the panchromatic image while maintain the spectral characteristics of the multi-spectral image well,and it outperforms several state of the arts.As for running time,it has rather obvious advantage over RKHS method.On the other hand,this thesis takes a classical framework(the unified framework of component substitution and multi-resolution analysis)into account,then integrates it with an RoG based filtering method which is originally used for image smoothing,and proposes another new pan-sharpening algorithm.In the algorithm,this filtering method is adopted both to extract spatial details from the panchromatic image and to modify the upsampled multi-spectral image so that it matches the details extracted better.Numerical experiments demonstrate that this algorithm is able to gain very competitive results compared with several popular algorithms which are subject to the classical framework as well. |
