Multi-modality Medical Image Fusion Algorithms Based On NSST

Medical image fusion technology effectively solves the limitations of single modality medical image for human tissue and organ information,and improves the utilization efficiency of medical image information,which has important theoretical research significance and application value for medical clinical diagnosis.Over the years,image fusion technology is developing rapidly,many advanced research results have been widely applied to the field of medical image fusion.In this thesis,according to the characteristics of medical images and the subgraph coefficients under different scales and directions in multi-scale transformation,some research on medical image fusion algorithms under the framework of multi-scale decomposition was finished,The main research work is as follows:1)In order to solve the limitation of single modality of medical images,this thesis combines with the feature that Non-Subsampled Shearlet Transform(NSST)can capture the details of an image,a multimodal medical images fusion algorithm based on NSST and improved PCNN is proposed.First,the source medical images are decomposed into low and high frequency subbands by NSST.Moreover,regional characteristics of the low frequency subband coefficients are obtained from regional energy and variance,a fusion rule based on weighted regional characteristics is adopted for the low frequency subband coefficients;and with the firing times and average gradient weighted method,the coefficients of the inner high frequency bands of medical images after decomposition were fused,and the firing times are determined by the improved pulse coupled neural networks(PCNN);the coefficients of the outer high frequency bands were combined by the maximal regional absolute value.Finally,the inverse NSST is used to produce the fused image.2)According to characteristics of complexity and diversity of medical image,combining the theory of multi-scale geometric analysis with the single scale sparse representation,a novel algorithm with sparse representation and non-subsampled Shearlet transform(NSST)for medical image fusion is proposed.Firstly,the NSST is used to decompose the registered source images,thus the low frequency sub-band coefficients and high frequency sub-bands coefficients can be obtained.Secondly,by K-SVD algorithm,the low frequency sub-band coefficients with lower sparseness are used to train the overcomplete dictionary and the sparse representation coefficients are calculated by the trained overcomplete dictionary,and fused according to the sparseness,energy and gradient characteristics of sparse representation coefficients.And then,with the sum of modified laplacian and average gradient weighted method,the high frequency coefficients with higher sparseness of medical images were fused.Finally,low and high frequency images are fused by inverse transformation of NSST to get the final image.3)In this thesis,lots of fusion experiments on gray images and color images are carried out respectively,and compared with the quality of the fusion images produced by different fusion algorithms.The experimental results show that the proposed algorithm effectively increases the complementary information of different modality medical image.The fusion image information is more abundant,and the detail information such as edges and textures is clearer,which can achieve good visual effects,and it can help doctors analyze the disease.At the same time,by analyzing image quality evaluation indexes objectively,we can see that the edge preservation index and structural similarity of the proposed algorithm's fusion image has achieved the best,standard deviation and spatial frequency is good,indicating that the fusion image is richer edge information and it can maintain better image structure,more scattered gray scale distribution.