Algorithm Research Based On Neural Network Image Fusion

Image fusion belongs to an important branch in the field of image processing.It mainly uses multiple source image data information of the same object or scene collected by different sensors to pass the complementary information of each source image or people through a specific fusion algorithm.The required information is extracted to obtain more accurate,richer and more comprehensive image information.Nowadays,the technology has been widely used in industry,medicine and aerospace.At present,there are many methods for image fusion technology,such as weighted average method,multi-scale transformation method and neural network.Among them,the neural network has a good biological background,and the global characteristics of the image can be considered when performing image processing.Pulse Coupled Neural Network(PCNN)is a third-generation artificial neural network.In addition to its good biological background,it also has the characteristics of pulse modulation and coupling.It is now widely used in the field of image fusion.And achieved certain results.However,in the traditional PCNN model,there are many parameters,and most of the parameter values are fixed values that are set by experience.In addition,the traditional PCNN model is more complex and computationally intensive,so how to improve the model and better set the parameters is an important research direction.Therefore,in view of these problems,this paper improves from two aspects.One is to improve the original PCNN single-channel model and expand it into two channels by improving the model.The other is combined with other methods.This paper combines PCNN with Non-Subsampled Shearlet Transform(NSST)for image fusion.The following is the main research work and content of this paper:(1)Explain the basic knowledge and working principle of PCNN,then introduce the classic algorithm of dual-channel PCNN image fusion.In view of the shortcomings of this model,this paper proposes an improved dual-channel PCNN model.Firstly,it is improved and simplified according to the characteristics of the PCNN model.In terms of the input of the model,the traditional spatial gray value is replaced by the improved spatial frequency.In terms of the link coefficient,Laplacian energy is used instead of the traditional one.The fixed value is then input into the model to obtain the final fusion result.(2)Combining the PCNN method with NS ST,this paper makes full use of the advantages of NS ST and PCNN in processing image fusion,and proposes the fusion method of PCNN in NSST domain.Firstly,the source image is decomposed into low frequency and high frequency;by NSST transform,and the high frequency subband is processed by PCNN method.The values of input and link coefficients are set to be adaptive,and the input space still uses the improved spatial frequency.The coupling link coefficient is set to image sharpness.The low-frequency sub-band is processed by the method of regional energy sum.The processed high frequency and low frequency sub-bands are inversely transformed into NSST to obtain a fused image.Finally,the effects of the two improved fusion methods are compared with those of other traditional fusion methods.It is concluded that the method of this paper is superior in terms of subjective performance and objective performance.