Image Fusion Algorithm Based On Sparse Representation And Non-subsampled Shearlet Transform

Image fusion technology is a process of fusing the important information of multiple images from different sensors to an image.The fused image integrates the complementary information of multiple source images and removes the redundant information in the source image,so the fused image can describe the scene more accurately and contains more information.The fused image contains more information and provides reliable source images for further processing.Therefore,in order to improve the efficiency of fusion and the efficiency of fusion,in this thesis,sparse representation technology is introduced into the traditional fusion algorithm.Firstly,combining the characteristics of multi-focus image with the advantages of guide filtering,in this thesis,a multi-focus image fusion algorithm based on sparse representation and guided filtering(Sparse Representation-Guided Filter,SR-GF)is presented.This algorithm not only has the advantages of dimensionality reduction and compression of sparse representation,but also has the advantage of preserving the edge of guide filtering.The fused image is perfect in space continuity and the fusion effect is improved on the objective index.Secondly,a image fusion algorithm in the transform domain based on sparse representation(SR)and guided filtering(GF)is studied in this thesis.The source image is transformed by NSST,high frequency signal and low frequency signal are obtained.High frequency signal fusion rule is based on sparse representation.Low frequency signal is filtered to get a completely blurred image and an residual image.The fuzzy image uses the fusion method of mean value and the residual image uses the fusion algorithm based on the guide filter.The fused image remains perfect in space continuity,and the artificial texture introduced is less.Finally,in this thesis,a novel image fusion method is proposed,which combines residual image(RD)with sparse representation.The source image is transformed by NSST,high frequency signal and low frequency signal are obtained.The low frequency signal is fused based on improved sparse representation fusion rules.The high frequency signal is fused based on residual image.The whole fusion algorithm not only has the advantage of sparse representation,but also has the adaptive advantage of PCNN,which preserves the detail information in the residual image.