Research On Medical Image Fusion Algorithms Based On Non-subsampled Shearlet Transform

Multimodality medical image fusion technology can obtain comprehensive and clear medical images by extracting and integrating complementary information from different modality medical images,which helps to improve the accuracy and reliability of clinical diagnosis.As a new multiscale geometric analysis tool,non-subsampled shearlet transform achieves efficient nonlinear approximation of the two-dimensional image.In order to effectively represent and analyse medical images and transfer more features from source images to fused image,non-subsampled shearlet transform combined with image processing techniques,such as morphological component analysis,moving frame decomposition,structure tensor and edge-preserving smoothing filter,are studied for the fusion of medical images.The main research contents are summerized as follows:Considering the low-pass subband decomposed with non-subsampled shearlet transform from source image still contains abundant features,a medical image fusion algorithm based on morphological component analysis and non-subsampled shearlet transform is proposed.Morphological component analysis is performed on the low-pass subbands and then the separated cartoon components and texture components are fused respectively.The sum-modified-Laplacian is used to select high-pass subband coefficients in order to extract more information from source image.The final fused image is obtained by using inverse non-subsampled shearlet transform on fused subbands.The experimental results show that the fused image obtained from this method is superior in both the visual effect and objective evaluation.In order to extract and transfer more features from source image to the fused image,a medical image fusion algorithm based on a moving frame decomposition framework and non-subsampled shearlet transform is proposed.Firstly,the source image is decomposed into texture component and approximate component based on moving frame decomposition framework.The texture components are fused using maximum fusion rule to retain salient gradients information,and the method based on non-subsampled shearlet transform is used for the fusion of approximation components.The final fused image is obtained by the proposed components synthesis process.Experimental results show that more details informtion are retained in fused image.Take advantage of structure tensor in capturing geometric features of image,a medical image fusion algorithm based on structure tensor and non-subsampled shearlet transform is proposed.First of all,the pre-fused gradient information is obtained from weighted structure tensor and the pre-fused coefficients are generated by fusing non-subsampled shearlet transform coefficients.Then the medical image fusion is converted into an optimization problem,which simultaneously constrains the gradient and the coefficient of fused image.The final fused image can be obtained by solving the constructed optimization problem.The experimental results show that the proposed algorithm has better fusion performance.Edge-preserving smoothing filter can preserve edge information while smoothing image,therefore a medical image fusion algorithm based on gradient minimization edge-preserving smoothing filter and pulse coupled neural network is proposed.First of all,a multiscale edge-preserving decomposition framework is proposed to decompose the source image into a base image and a series of detail images.Then regional weighted sum of energy is proposed to fuse base images so as to avoid the reduction of contrast and loss of information,and pulse coupled neural network is used for the fusion of detail images in order to transfer more detail information from source image into the fused image.Finally,the final fused image is obtained by synthesising the fused base and detail images.The experimental results show that the fused image obtained from proposed algorithm can effectively retain the edge information and main features.In order to efficiently capture the details information in the medical image,a medical image fusion algorithm based on gradient minimization edge-preserving smoothing filter and shearing filter is proposed.Utilizing gradient minimization smoothing filter and Gaussian low-pass filter,a multiscale joint decomposition framework is constructed.Source image is decomposed into low-frequency component images,edge feature images,and detail images at multiple scales using proposed multiscale joint decomposition framework.The detail-enhanced image is obtained by calculating the weighted sum of edge feature image and detail image,and shearing filter is used to extract the geometric feature information in detail-enhanced image.Finally,the fused image is obtained by synthesising low-frequency image and directional coefficients.The experimental results verify that the proposed algorithm is effective in the fusion of multimodality medical images,and the performance is better than the compared methods.