GFF导向滤波器的图像融合的MATLAB仿真

function [F] = GFF(I,r1,eps1,r2,eps2) % This is the MATLAB implementation of the paper: S.Li,Xudong Kang, and % J. Hu, "Image Fusion with Guided filtering", IEEE Transactions on Image % Processing, 2013. % Example: % % I(:,:,1)=imread('source1.tif'); % I(:,:,2)=imread('source2.tif'); % .... % I(:,:,N)=imread('sourceN.tif'); % F=GFF(I); % The default parameter is set to be: % r1=45,eps1=0.3,r2=7,eps2=10^-6 % NOTE: This code is provided for research purposes only. It comes with no % warranty or guarantee of any kind. If you use this code, please cite our % paper presented above. Only the code for gray scale image fusion is % presented here. % % The code is edited by Xudong Kang in 01,Jan.,2013 if ~exist('r1') r1 = 45; eps1 = 0.3; r2 = 7; eps2 = 10^-6; else end if size(I,3)==3 && size(I,4)~=1 G=rgb2gray_n(I); else G=I; end % Laplician Filtering H = LapFilter(G); % Gaussian Filtering S = GauSaliency(H); % Initial Weight Construction P = IWconstruct(S); % Weight Optimization with Guided Filtering W_B = GuidOptimize(G,P,r1,eps1); W_D = GuidOptimize(G,P,r2,eps2); % Two Scale Decomposition and Fusion F = GuidFuse(I,W_B,W_D); % Image Format Transformation F = uint8(F*255); %imwrite(F,'fused_gff.tif') function [G] = rgb2gray_n( I ) N=size(I,4); G=zeros(size(I,1),size(I,2),N); for i=1:N G(:,:,i)=rgb2gray(I(:,:,:,i)); end function [ H ] = LapFilter( G ) %Conduct Laplacian filtering on each source image L = [0 1 0; 1 -4 1; 0 1 0]; % The 3*3 laplacian filter N = size(G,3); G = double(G)/255; H = zeros(size(G,1),size(G,2),N); % Assign memory for i = 1:N H(:,:,i) = abs(imfilter(G(:,:,i),L,'replicate')); end function [ S ] = GauSaliency( H ) % Using the local average of the absolute value of H to construct the % saliency maps N = size(H,3); S = zeros(size(H,1),size(H,2),N); for i=1:N se = fspecial('gaussian',11,5); S(:,:,i) = imfilter(H(:,:,i),se,'replicate'); end S = S + 1e-12; %avoids division by zero S = S./repmat(sum(S,3),[1 1 N]);%Normalize the saliences in to [0-1] function [P] = IWconstruct( S ) % construct the initial weight maps [r c N] = size(S); [X Labels] = max(S,[],3); % find the labels of the maximum clear X for i = 1:N mono = zeros(r,c); mono(Labels==i) = 1; P(:,:,i) = mono; end function [W] = GuidOptimize( I, P, r, eps) N = size(I,3); I = double(I)/255; for i=1:N P(:,:,i) = double(P(:,:,i)); W(:,:,i) = guidedfilter(I(:,:,i), P(:,:,i), r, eps); end W = uint8(W.*255); % Remove values which are not in the [0-1] range W = double(W)/255; W = W + 1e-12; %Normalization W = W./repmat(sum(W,3),[1 1 N]); function [F] = GuidFuse(I, W_B, W_D) I = double(I)/255; se = fspecial('average', [31 31]); if size(I,3)==3 && size(I,4)~=1 [r,c,M,N]=size(I); F_B = zeros(r,c,M); F_D = zeros(r,c,M); for n=1:N w_B = W_B(:,:,n); w_D = W_D(:,:,n); G = I(:,:,:,n); B = imfilter(G,se,'replicate'); D = G-B; F_B = F_B+B.*repmat(w_B,[1 1 3]); F_D = F_D+D.*repmat(w_D,[1 1 3]); end else [r,c,N] = size(I); F_B = zeros(r,c); F_D = zeros(r,c); for n=1:N w_B = W_B(:,:,n); w_D = W_D(:,:,n); B = imfilter(I(:,:,n),se,'replicate'); D = I(:,:,n)-B; F_B = F_B+B.*w_B; F_D = F_D+D.*w_D; end end F = F_B+F_D;