Research On Multi-modal Medical Image Fusion Algorithm Based On Nosubsampled Shearlet Transform

With the rapid development of sensing technology,the image imaging system is gradually improved,and the image information obtained is growing explosively,these information have both complementarity and redundancy.Image fusion technology can fuse multiple images into one image by different types of sensors or the complementary information of the same type of sensors,and generate a new image after eliminating redundant information.The fused image can make up for the deficiency of a single image,and can improve the utilization of image information,obtain more rich and more accurate information,and use the acquired information to generate high-quality images.Image fusion technology has been widely used in medical imaging,military,remote sensing,digital imaging and other fields.This paper will mainly focus on the research of multi-modal medical image fusion algorithm.The main contents are summarized as follows:1.In view of the problem that parameters in Pulse Coupled Neural Network(PCNN)need to be manually set according to experience or experimental results,a method is proposed to set the Improved Sum Modified Laplacian(ISML)as the connection strength of PCNN.Firstly,two images are decomposed by using the Nonsubsampled Shearlet Transform(NSST),and the source image is decomposed into one low-frequency subband and several highfrequency subbands.The original Weighted Local Energy(WLE)and Spatial Frequency(SF)were fused by using the maximum fusion strategy.However,when the energy and detail information of the pixels at the same location are close,this fusion strategy may not be suitable.Therefore,a method based on WLE and SF weighted average is proposed for energy preservation and detail extraction of low-frequency subbands.Secondly,the traditional PCNN model raqureires manual setting the values of parameters,such as connection strength,amplitude,attenuation coefficient,etc.Among which connection strength is the most important parameter.Moreover,the traditional PCNN ignores the correlation between pixels.In order to realize the adaptive setting of connection strength and consider the information of neighborhood pixels,ISML is set as connection strength in the PCNN model.Then,the improved PCNN model was calculated to obtain the firing times,and the fusion coefficient of the high frequency subbands was determined according to the firing times.Finally,the inverse transform of NSST is used to reconstruct the fused low frequency subbands and high frequency subbands to obtain the final fused image.In this paper,experiments are carried out on medical brain images of different modes.The results show that the improved method achieves good visual effects and objective evaluation results.2.A Generalized Random Walk(GRW)combined with eight-neighborhood SML image fusion model is proposed to solve the ambiguity and artifact problems in image fusion methods.Different from most previous fusion methods,image fusion is regarded as a probabilistic synthesis process.In the case of considering neighborhood information,GRW is used to obtain the final probability through global optimization.These probability graphs are used as weights in the fusion process to generate the fusion image.In the probability diagram,the higher the pixel energy,the higher the probability.First,the two source images are decomposed by NSST.Then,the low frequency subbands are compared to obtain the initial decision image,and the edges of the decision image are smoothed by the multi-scale guide filter.Finally,GRW is added to obtain the final decision image.The fused low frequency subbands are obtained by combining the obtained decision graph with the low frequency subbands.High frequency subbands contain a lot of contour information,and the original SML contains only vertical and horizontal information,which is easy to be lost during information extraction.Therefore,the SML is extended from two neighborhoods to eight neighborhoods.By calculating the eight-neighborhood SML of the pixel,the high-frequency subbands with large values are selected as the high-frequency subband coefficients after fusion.Finally,the inverse transform of NSST is used for reconstruction to obtain the final fused image.The experimental results show that compared with the five medical image fusion methods,the proposed method can effectively avoid blurring and artifacts,This method achieves more competitive performance and improves the visual quality and objective evaluation indexes.