Research On Techniques Of Medical Image Fusion Based On Multiresolution Analysis

Medical image fusion is the process of synthesizing medical images of different modalities and synthesizing a fused image containing important information of all the source images,so as to obtain a more complete diagnosis conclusion than single modality medical images.In order to be able to analyze the detailed information of each different resolution in the image,the multi-resolution analysis method has become a well-accepted fusion method of performance in the field of image processing.In particular,there are two theories of non-subsampled contourlet transform(NSCT)and nonsubsampled shearlet transform(NSST)with translational invariance,which have excellent detail capture capability.This thesis will focus on medical image fusion methods based on NSCT and NSST.The main research contents are summarized as follows:Firstly,several typical image fusion methods based on multi-resolution analysis are studied,including pyramid transform,wavelet transform and contourlet transform.Although the method based on the pyramid transformation is simple in implementation,the result is easy to appear as a block-like mark.Wavelet transform has good time-frequency local analysis ability,but there are great limitations in dealing with two-dimensional image signals.Contourlet transform method has higher direction sensitivity and better non-linear approximation performance.Subsequent research of this paper is based on contourlet transform.Then,an image fusion algorithm based on nonsubsampled contourlet transform(NSCT)and improved pulse neural network(PCNN)is proposed.NSCT not only inherits the multi-scale and good time-frequency characteristics of wavelet transform,but also has the invariance of translation.Because the traditional PCNN model has more pending parameters and slightly more computational complexity,this paper simplifies the original model by using Unit-linking PCNN,and replaces the dynamic threshold with a linear function that decreases monotonously with the number of iterations.The linking strength parameter is set to the pixel resolution of the region,we designed an image fusion method based on NSCT domain and improved PCNN.The simulation results show the effectiveness of the proposed method.Finally,aiming at the problem of gray-scale and color medical image fusion,an image fusion method based on NSST and non-negative matrix factorization(NMF)model is proposed.The NSST tool has a more flexible structure for decomposing images.There are some improvements in the rules of low-frequency and high-frequency sub-band fusion.First,it is found that the mathematical structure of singular value decomposition(SVD)is very similar to that of NMF model.Therefore,using SVD to construct nonnegative matrix,The second one is to combine the advantages of visual sensitivity coefficient and energy matching degree comprehensively and complementarily so as to effectively distinguish the background information from the target area.Finally,MRI-SPECT and MRI-PET images are taken as experimental objects.The proposed algorithm can preserve more color information components and meet the human vision's understanding and perception of image content.