多聚焦圖像融合源代碼.doc

針對(duì)經(jīng)典的最大系數(shù)法不準(zhǔn)確和方差法計(jì)算量大的問(wèn)題，本文給出了一種混合多級(jí)式多聚焦圖像融合方法。對(duì)于三層小波分解的多聚焦圖像融合，每幅圖像被分解為三層十個(gè)頻帶。對(duì)這十個(gè)頻帶本文分別采用三種方法進(jìn)行融合。對(duì)于低頻系數(shù)，本文仍然采用求平均法；對(duì)于高頻系數(shù)本文采用方差法和最大系數(shù)法進(jìn)行融合。它們的計(jì)算量比最大系數(shù)法大一些，但是融合結(jié)果更接近于原始清晰圖像，而相比于方差法，它們的計(jì)算量小的多，但是融合質(zhì)量稍差一些，應(yīng)用者可以根據(jù)不同的需要進(jìn)行選擇。本文還給出了一種基于Canny算子邊緣檢測(cè)的小波變換多聚焦圖像融合方法。首先對(duì)圖像進(jìn)行三層小波分解，然后用Canny算子進(jìn)行邊緣檢測(cè)，得到各層分辨率下的邊緣圖像；對(duì)相應(yīng)分辨率的高頻小波系數(shù)根據(jù)其是否為圖像的邊緣點(diǎn)采用最大系數(shù)法或方差法分別進(jìn)行融合。仿真實(shí)驗(yàn)證明該方法效果良好，計(jì)算量可以靈活調(diào)節(jié)。關(guān)鍵詞:小波變換；多尺度幾何分析；多聚焦圖像融合；邊緣檢測(cè)主要程序：clear all;close all;leo1=imread(a1.bmp);%讀入圖片leo2=imread(a2.bmp)T=0.4;k1=0.5;k2=0.5;w=db4;m=edge; tic;outdoor1=leo1;outdoor2=leo2;%三層小波分解ca11,chd11,cvd11,cdd11=dwt2(outdoor1,w);ca12,chd12,cvd12,cdd12=dwt2(ca11,w);ca13,chd13,cvd13,cdd13=dwt2(ca12,w);ca21,chd21,cvd21,cdd21=dwt2(outdoor2,w);ca22,chd22,cvd22,cdd22=dwt2(ca21,w);ca23,chd23,cvd23,cdd23=dwt2(ca22,w);%求邊緣圖像e11=edge(ca11,canny,T);e12=edge(ca12,canny,T);e13=edge(ca13,canny,T);e21=edge(ca21,canny,T);e22=edge(ca22,canny,T);e23=edge(ca23,canny,T);%矩陣融合chd3=matfusion(chd13,chd23,e13,e23);cvd3=matfusion(cvd13,cvd23,e13,e23);cdd3=matfusion(cdd13,cdd23,e13,e23);chd2=matfusion(chd12,chd22,e12,e22);cvd2=matfusion(cvd12,cvd22,e12,e22);cdd2=matfusion(cdd12,cdd22,e12,e22);chd1=matfusion(chd11,chd21,e11,e21);cvd1=matfusion(cvd11,cvd21,e11,e21);cdd1=matfusion(cdd11,cdd21,e11,e21);ca3=k1*ca13+k2*ca23;%反小波變換L2=size(chd2);L1=size(chd1);ca2=idwt2(ca3,chd3,cvd3,cdd3,w);ca1=idwt2(ca2(1:L2(1),1:L2(2),chd2,cvd2,cdd2,w);I=idwt2(ca1(1:L1(1),1:L1(2),chd1,cvd1,cdd1,w);toc;%顯示圖像 subplot(2,2,1);imshow(leo1);colormap(gray);title(input1);axis square subplot(2,2,2);imshow(leo2);colormap(gray);title(input2);axis square subplot(2,2,3);imshow(uint8(I);colormap(gray);title(最大系數(shù)法與方差法融合);axis square; %標(biāo)準(zhǔn)差 f_stdvar=sta_variance1(I) %清晰度，又稱為平均梯度 f_avggrad=avg_gradient(I) f=PSNR(I)% 信噪比 %其值越大，說(shuō)明融合圖像和標(biāo)準(zhǔn)參考圖像%的差異越小，融合效果越好se=RMSE(I)% %RMSE(均方根誤差,也稱%為標(biāo)準(zhǔn)偏差， 融合圖像與標(biāo)準(zhǔn)圖像差異%程度，差異越小說(shuō)明融合圖像與標(biāo)準(zhǔn)圖%像越接近%I3 = imread(clockB.jpg);sI,cI,bI = size(uint8(I);I3 = double(uint8(I)+1;s(256) = 0;for i = 1:sI for j = 1:cI va = I3(i,j); s(va) = s(va)+1; endendp = s./(sI*cI);en = 0.;for i = 1:256 if p(i)=0 en = en+p(i)*log2(p(i); endendentropy = -en %融合圖像的熵%function f=PSNR(h2)%PSNR return PSNR(峰值信噪比) 能用 %衡量融合圖像相對(duì)于標(biāo)準(zhǔn)參考圖像灰度%的偏離程度其值越大，說(shuō)明融合圖像和%標(biāo)準(zhǔn)參考圖像的差異越小，融合效果越好h1=imread(a.bmp) ;% 標(biāo)準(zhǔn)圖像 G1=double(h1);G2=double(h2);m1,n1=size(G1);m2,n2=size(G2);m=min(m1,m2);n=min(n1,n2);b=0;b=max(G1(:);%一般是255c=0;for i=1:m for j=1:n w=G1(i,j)-G2(i,j); c=c+w2; endendf=10*log10(b2*m*n/c);%function se=RMSE(h2)%RMSE return RMSE(均方根誤差) 能用 %融合圖像與標(biāo)準(zhǔn)圖像差異程度，差異越小%說(shuō)明融合圖像與標(biāo)準(zhǔn)圖像越接近h1=imread(a.bmp) ; % 標(biāo)準(zhǔn)圖像 G1=double(h1);G2=double(h2);m1,n1=size(G1);m2,n2=size(G2);m=min(m1,m2);n=min(n1,n2);c=0;for i=1:m for j=1:n w=G1(i,j)-G2(i,j); c=c+w2; endendse=sqrt(c/(m*n);%標(biāo)準(zhǔn)差function img_var = sta_variance1(img);if nargin=1 img = double(img); r,c,b = size(img); for k = 1 : b band = img(:,:,k); s1=sum(sum(band.2)./(r*c); s2=(sum(sum(band)./(r*c).2; g(k) = s1-s2; end img_mean = mean(g); img_var=sqrt(img_mean);else disp(wrong number of input);End%平均梯度，也稱清晰度f(wàn)unction outval = avg_gradient(img)if nargin = 1 img = double(img); r,c,b = size(img); dx = 1; dy = 1; for k = 1 : b band = img(:,:,k); dzdx=0.0; dzdy=0.0; dzdx,dzdy = gradient(band,d