Research On Multi-Source Image Fusion Algorithm Based On Multi-statistic Properties And Adaptive DCPCNN In NSST Domain

With the advancement of science and technology,the resolution of optical equipment imaging systems is becoming higher and higher,but due to the limited imaging range of existing imaging equipment,it is difficult to achieve that all objects in the same image are clear.Therefore,multi-source image fusion technology has been proposed by scholars as far as possible to ensure that all objects in the same image are clear,and this method has been rapidly developed in many fields.Firstly,this master's thesis gives a detailed introduction to some of the image fusion,such as background,significance,domestic and international research status,etc.Secondly,this master's thesis introduces several traditional multi-source image fusion methods in frequency domain and time domain,as well as two subjective and objective evaluation methods.Again,this master's thesis uses discrete wavelet transform(DWT),Laplacian pyramid transform(LPT),and non-subsampled contourlet transform(NSCT),and non-subsampled shearlet transform(NSST)four multi-scale transform methods respectively to fuse multiple groups of images.Through many experimental comparisons and analysis,the superiority of the NSST algorithm is verified.Image fusion experiments based on pulse coupled neural network(PCNN),simplified pulse coupled neural network(S-PCNN),and dual channel adaptive pulse coupled neural network(DCAPCNN)are performed respectively to verify the superiority of the NSST algorithm.Finally,based on above algorithms,a multi-source image fusion algorithm combining multi-statistical features and adaptive DCPCNN in NSST domain is proposed.Local variance,new sum of modified Laplacian(NSML)and modified spatial frequency(MSF)three statistical characteristics are weighted and averaged to be used as a fusion rule of adaptive DCPCNN's link adaptive selection input in low-frequency.High-frequency fusion rules use phase consistency,the NSST inverse transform is used to obtain the fusion effect graph.Experiments are compared with the existing algorithms,and the experimental results and data are analyzed both subjective and objective aspects,which verify the effectiveness of the proposed algorithm.