Research On The Medical Image Fusion Algorithm Based On NSST

With the rapid development of sensing technology since the middle of the last century,many medical image imaging systems with different functions have emerged and have been widely used in the field of medicine,however,each medical image has its own advantages and limitations,so doctors need a variety of medical images to provide more information about the lesion,and medical image fusion technology emerges.This thesis focuses on the research of medical image fusion algorithm based on NSST(Non-Subsampled Shearlet Transform),and proposes two methods of medical image fusion based on NSST.One is the medical image fusion algorithm based on NSST and DWT(Discrete Wavelet Transform);the other is the medical image fusion algorithm based on NSST and improved adaptive PCNN(Pulse-coupled Neural Network).The following are the main work and innovations for this thesis:1.Proposed medical image fusion algorithm based on NSST and DWT.We prove by experiments: The NSST transform has a strong ability to express the edge and texture(singularity features of the line)of image,but it is relatively weak to express the singularity feature of the point.However,the DWT transform has a strong ability to express the singularity feature of the point in the image,but it is relatively weak to express the singularity feature of the line.According to the complementary characteristics of these two multi-scale decomposition methods,and great differences and complementarities between multimodal medical images,so this thesis proposes a decomposition method based on NSST and DWT,which is a medical image fusion algorithm based on NSST and DWT.Compared with other single transformation tools,this algorithm not only has a great performance improvement in objective image evaluation index,but also has excellent performance in subjective vision.2.Improved traditional SML(Sum-Modified Laplacian)fusion rules.The traditional SML without considering the energy information of the diagonal directions only contains the Laplace energy of the horizontal and vertical directions,which means that using traditional SML as the fusion rule of fusion image may lose important information,so this thesis proposes a rule of image fusion based on ISML(Improved Sum-Modified Laplacian).3.Proposed a medical image fusion algorithm based on NSST and improved adaptive PCNN.Traditional image fusion methods based on spatial domain or transform domain are easy to be affected by noise during the fusion process,and do not make effective use of the global features of the image,which leads to local brightness and structural profile information of the fused image not harmonious.Considering that the contour information of CT and MRI images occupies most of the area of the entire image,the contour information is mainly included in the low frequency sub-band coefficients,in addition,PCNN has an advantage in processing image global feature fusion.In this thesis,a medical image fusion algorithm based on NSST and improved adaptive PCNN is proposed.The experiments verify the superiority of ISML and local structural information factors as input and link strength of PCNN,respectively,so as to achieve an adaptive pulse coupled neural network.The sum of PCNN fire amplitude is used as a selection criterion for fusing image coefficients,which greatly reduces the number of iterations of PCNN.Compared with the algorithm based on PCNN in recent years,both the subjective and objective performance of the proposed algorithm have been greatly improved.