基于鲁棒性的数字水印的嵌入与提取

数字水印技术(Digital Watermarking)是通过一定的算法将一些标志性信息直接嵌入到多媒体内容当中，但不影响原来内容的价值和使用，并且不能被人的感知系统察觉或者注意到，只有通过专门的检测器或者阅读器才能提取。如图1所示：左上角是要嵌入的水印，右上角是宿主图像，左下角是嵌入水印之后的图像。

图1   (1)水印  (2)原始宿主图像 (3)嵌入水印之后的宿主图像

数字水印常见的分类：

（1）    根据数字水印是否可见分为：可见水印、不可见水印。

（2）    根据数字水印的作用可以分为：鲁棒性水印、脆弱性水印、半脆弱性水印。

（3）    根据水印实现的方法分为：时（空）域水印、频域数字水印。

数字水印系统的组成：嵌入和提取。

数字水印嵌入的一般过程如图2所示：

图2 数字水印嵌入的一般过程

数字水印提取的一般过程如图3所示：

图3 数字水印提取的一般方法

基于DCT的的鲁棒性水印嵌入过程如图4所示：

图4 基于DCT的的鲁棒性水印嵌入

 

 

 

水印的提取步骤

function [snr , watermarked_image_int] = water_mark1(coypright,cover_object,k,blocksize)if nargin<4    coypright=[ 'coypright.bmp'];  %水印    cover_object = ['len_std__.jpg'];%原图片    k = 20;%设置水印强度    blocksize = 8;  %设置图像的分块大小为blocksize*blocksize；endmidband = [0 0 0 1 1 1 1 0 ;           0 0 1 1 1 1 0 0 ;           0 1 1 1 1 0 0 0 ;           1 1 1 1 0 0 0 0 ;           1 1 1 0 0 0 0 0 ;           1 1 0 0 0 0 0 0 ;           1 0 0 0 0 0 0 0 ;           0 0 0 0 0 0 0 0 ;];message = imread(coypright);%为什么要将读入的图像转换为双精度呢subplot(2,2,1);imshow(message);%显示水印[Mm,Nm] = size(message);%计算水印的大小   Mm=24;Nm=64;n = Mm*Nm;message = reshape(message,1,n);cover_object =  imread(cover_object)  ;%读入宿主图片并且将其转换成双精度Mc = size(cover_object,1);  Nc = size(cover_object,2);c = Mc/blocksize;   d = Nc/blocksize;   m=c*d;%计算图像划分的图像块%计算宿主图像每一块的方差xx =1;for i=1:c    for j=1:d       pjhd(xx) = sum( sum( cover_object(1+(i-1)*8:i*8 ,1+(j-1)*8:j*8 )))/64;       fc(xx) = sum( sum( (cover_object(1+(i-1)*8:i*8 ,1+(j-1)*8:j*8 )- pjhd(xx)).^2))/64 ;       xx = xx+1;    end      end%取出方差最大的前n块   [A,B] = sort(fc);    %B = A( (c*d-n+1) : (c*d) );%因为A中是从小到大排序了p = B( (c*d-n+1) : (c*d) );%因为A中是从小到大排序了%将水印信息嵌入到方差最大的前n块 fc_o = ones(1,c*d);   %1是白色的 fc_o(p) = message;message_vetot = fc_o;watermarked_image = cover_object;rand('state',7);%设置随机数生成器状态，作为系统密钥%根据当前的随机数生成器生成 0 1 的伪随机数pn_sequence_zero = round(rand(1,sum(sum(midband))));%嵌入水印x = 1;y=1;for kk=1:m    %分块DCT变换    a = cover_object(y:y+blocksize-1,x:x+blocksize-1);    dct_block = dct2(a);    ll=1;    if message_vetot(kk) == 0   %这个块里应该嵌入水印        for ii=1:blocksize            for jj=1:blocksize                if(midband(jj,ii) == 1)                    dct_block(jj,ii) = dct_block(jj,ii)+k*pn_sequence_zero(ll);                    ll = ll+1;                end            end        end    end        %分块DCT反变换    watermarked_image(y:y+blocksize-1,x:x+blocksize-1) = round(idct2(dct_block));    b = idct2(dct_block);    %换行    if x+blocksize>=Nc        x = 1;y = y+blocksize;    else        x = x+blocksize;    endendwatermarked_image_int = uint8(watermarked_image);%生成并输出嵌入水印后的图像imwrite(watermarked_image_int,'dct2_watermarked.bmp','bmp');subplot(2,2,2);imshow(cover_object);%显示原来图像 subplot(2,2,3);imshow(watermarked_image_int,[]);%显示嵌入后的图像%显示峰值信噪比xsz = 255*255*Mc*Nc/sum ( sum( (cover_object-watermarked_image).^2 ));psnr = 10*log10(xsz);snr = psnr(:,:,1); fprintf('信噪比是 %f',snr);end

function [sim ,message]=water_mark2(cover_object,watermarked_image,orig_watermark,blocksize)if nargin<4    cover_object = ['len_std__.jpg'];    watermarked_image =['dct2_watermarked.bmp'];    orig_watermark = ['coypright.bmp'];    blocksize = 8;endmidband = [0 0 0 1 1 1 1 0 ;    0 0 1 1 1 1 0 0 ;    0 1 1 1 1 0 0 0 ;    1 1 1 1 0 0 0 0 ;    1 1 1 0 0 0 0 0 ;    1 1 0 0 0 0 0 0 ;    1 0 0 0 0 0 0 0 ;    0 0 0 0 0 0 0 0 ;];cover_object = imread(cover_object)  ;%读入原宿主图像watermarked_image = imread(watermarked_image);%读入待检测的图像Mw = size(watermarked_image,1);Nw = size(watermarked_image,2);c = Mw/blocksize;  d = Nw/blocksize;  m = c*d;orig_watermark = double(imread(orig_watermark));%读入水印图像Mo = size(orig_watermark,1);    No = size(orig_watermark,2);   n=Mo*No;rand('state',7);%设置随机数生成器状态，作为系统密钥%根据当前的随机数生成器生成 0 1 的伪随机数pn_sequence_zeros = round(rand(1,sum(sum(midband))));%提取水印x = 1;  y = 1;for kk=1:m    dct_block1 = dct2(watermarked_image(y:y+blocksize-1, x:x+blocksize-1));    dct_block2 = dct2(cover_object(y:y+blocksize-1, x:x+blocksize-1));    ll = 1;    for ii=1:blocksize        for jj=1:blocksize            if (midband(jj,ii) == 1)                sequence(ll) = dct_block1(jj,ii) - dct_block2(jj,ii);                ll = ll+1;            end        end    end    %计算两个序列的相关性    if sequence == 0  %没有嵌入信息，相关性就比较大        correlation(kk) = 1;    else        correlation(kk) = corr2(pn_sequence_zeros,sequence);    end        %换行    if x+blocksize>=Nw        x=1;  y=y+blocksize;    else        x = x+blocksize;    endend%相关性大于0.5没有嵌入内容 ，小于0.5则表示曾经被嵌入for kk=1:m    if correlation(kk) == 1   %相关性比较大        message_vector(kk) = 1;%没有嵌入信息    else        message_vector(kk)=0;  %被嵌入了  1很多难道都被嵌入了吗？？？？    endend%计算原始图像的方差xx =1;for i=1:c    for j=1:d        pjhd(xx) = sum( sum( cover_object(1+(i-1)*8:i*8 ,1+(j-1)*8:j*8 )))/64;        fc(xx) = sum( sum( (cover_object(1+(i-1)*8:i*8 ,1+(j-1)*8:j*8 )- pjhd(xx)).^2))/64 ;        xx = xx+1;    endend%取出方差最大的前n块A = sort(fc);  B=A((c*d-n+1):c*d);%根据原始图像方差最大的前n块将水印提取出来fc_o = ones(1,n); %1是白色的H2=[];for g=1:n    for h=1:c*d        if( B(g) == fc(h))            fc_o(g)= message_vector(h);            break;        end    endendmessage_vector = fc_o;%重组嵌入的图像message = reshape(message_vector(1:Mo*No),Mo,No);%计算提取的水印和原始水印的相似度sim = corr2(orig_watermark,message);%把水印信息保存为‘message.bmp’imwrite(message,'message.bmp','bmp');figure;subplot(2,1,1);imshow(orig_watermark);subplot(2,1,2);imshow(message)fprintf('提取的水印与原水印的相似度 %f',sim);% end