Medical Image Fusion Algorithm Based On Pulse Coupled Neural Network

With the advent of the information age,medical imaging technology has grown by leaps and bounds.Imaging technology is not only used as an assitant tool in the past,but also it participates in the entire process of medical diagnosis.Therefore,medical imaging plays an indelible role in the dia gnosis and treatment.However,on account of the limitations of imaging equipment,there are still some deficiencies in current medical imaging.For example,certain diseases are diagnosed with multiple images of different modalities.Image fusion technology is focused on this issue.At present,there are many research methods in the field of image fusion at home and abroad,such as neural network,multi-scale decomposition and weighted average fusion methods.The neural network has a strong biological background and can take into account the global characteristics of the image.A special type of neural network from the cat's visual cortex,which is called as Pulse Coupled Neural Network,has been successfully applied to image fusion and has achieved good achievements.The network is capable of understanding and analyzing image on the basis of mammalian optic nerve cortex mechanism.However,the existing medical image fusion algorithms based on Pulse Coupled Neural Network have two main problems: O n the one hand,a large number of parameters and iteration times are set by hand or by experience,how to make a large number of parameters and iteration times accurately set is also an urgent problem to be solved.O n the other hand,the mathematical model of the network is more comp lex and has higher computational complexity.How to simplify the model reasonably and reduces the parameters in the model is an important research direction.Aiming at the existing problems of the image fusion algorithm based on Pulse Coupled Neural Network,the thesis will discuss the relevant issue from the following aspects:(1)Pulse Coupled Neural Network is combined with other technologies,and the important parameters are set to be adaptive.A fusion algorithm based on Compressive Sensing and adaptive Pulse Coupled Neural Network in Non-subsampled Shearlet Transform domain is proposed.Firstly,the source image is decomposed into high and low frequencies by Non-subsampled Shearlet Transform;Secondly,the improved Pulse Coupled Neural Network is used to fuse the low frequency subband coefficients and the sum of difference of two squares is adopted as the input to motivate Pulse Coupled Neural Network.Also,the sum of directional gradients is utilized as the linking strength and the high-frequency subband coefficients with larger calculation are taken by the method of Compressive Sensing;Finally,a fusion image can be obtained by the Non-subsampled Shearlet inverse transform.The algorithm is superior to the comparison algorithm both in subjective and objective aspects.(2)Simplify the Pulse Coupled Neural Network model,this thesis adopts the simplified model of Pulse Coupled Neural Network which is called Linking Synaptic Computation Network to fusion image and the image fusion algorithm is proposed by using the connection item of Linking Synaptic Computing Network model.First ly,the two images are inputed into the Linking Synaptic Computing Network model respectively;Secondly,the L term is used instead of the ignition frequency in the traditional Pulse Coupled Neural Network as the output;Then,the iteration is terminated by the multi-pass operation;Finally,the pixels of the fused image are obtained by comparing the value of the connection item.Theoretical analysis and experimental results show that the proposed algorithm is simple and easy.It not only reduces the number of parameters to be determined,but also reduces the computational complexity.Besides it solves the problem that the number of iterations in the traditional modelis uncertain.(3)Based on the algorithm,the thesis designed and implemented an image fusion system,in which the specific function of each module are introduced in detail.The system is mainly divided into five modules: comparing different reconstruction methods of Compressive Sensing,image fusion process based on Compressed Sensing and Pulse Coupled Neural Network in Non-subsampled Shearlet domain,image fusion module I,Linking Synaptic Computation Network image fusion process based on Non-subsampled Contourlet Transform and image fusion module II.The image fusion module I adapts Compressive Sensing and adaptive Pulse Coupled Neural Networks in Non-subsampled Shearlet Transform domain that is proposed in this paper.The medical image fusion algorithm based on Linking Synaptic Computation Network proposed in this paper is used in the image fusion module II.