The Study Of Image Fusion Based On Non-sampled Shearlet Transform

Because wavelet transform can capture the information of images in the limited directions,can't effectively represent high-dimensional data,therefore,multi-scale geometric analysis(MGA)method is improved,updated and replaced.In this paper,NSST in multi-scale geometric transform is mainly studied.Firstly,the basic principle of shearlet transform is introduced,the properties are analyzed.After that,we found NSST has strict mathematical theory,it not only has the characteristics of other multi-scale tools,but also has translation invariance.It can capture arbitrary direction information,so as to achieve effective approximation of the singular signal.Then the coefficients of NSST are analyzed,we found that the coefficients have region correlation in the same sub-band,and have father-son correlation at different scales in the same direction.Finally,according to the characteristics of the coefficient distribution,two image fusion methods are proposed which based on the direction region matching and tree structure matching,hidden Markov tree model and direction region matching.This kind of methods comprehensive considered the two kinds of correlations,thus the defects of original methods that only considers a single coefficient in different scales and in different directions are broken.The specific fusion methods are as follows:(1)In this paper,a NSST remote sensing image fusion algorithm based on directional region matching and tree model matching is proposed.First,the source images are decomposed by NSST.Then the sparse representation of source images are realized,and the details are remained.Secondly,using the direction template to capture the direction region correlation of coefficients in the same sub-band.After that,through the establishment of binary tree structure to capture the correlation coefficient between the father-son coefficients.Finally,by using adaptive threshold P that obtained by variance method to adjust the above two kinds of correlation.So the algorithm considers the directional region correlation and the father-son correlation of NSST coefficients in the same time.The experimental results show that the proposed algorithm is superior to other traditional methods.(2)In this paper,a NSST image fusion algorithm which based on the combination of the hidden Markov model with the directional region matching degree is proposed.First of all,the source images are decomposed by NSST,then the sparse representation of source images are realized,and the details are remained.Secondly,we analyze the properties of NSST coefficients,then,through hidden Markov tree model to train the sub-bands,and using the state probability of root node and the transition probability to estimate the dependence between father-son coefficients,so as to calculate the probability of each coefficient whosestate is large.Finally,we added the direction region matching degree for coefficients in the same sub-band in the fusion rules.This can realize the considerations of directional region correlation and father-son correlation in the same time.Experiment results verify the effectiveness of the algorithm.