我的想法是先使用分解图像dwt2，然后应用dct2到 coefficients_approximation，对结果应用水印，然后重新组合图像。
final_image = idwt2( idct2 ( dct2 ( dwt2 (starting_image))))
但是每当我这样做时，我都会失去很多质量，我不知道为什么。这是代码，有什么想法吗？
clear all; clc;
% read lena img
lena = double(imread('lena.jpg'));
% read watermark img
%w = imread('mark.png');
load cookiebears.mat
% compure DWT on lena and watermark
[approximate_lena, horizontal_lena, vertical_lena, diagonal_lena] = dwt2(lena, 'haar');
% i have to save the image first and then open it in order to use it in
% dct2() omitting this step generate an error
imwrite(uint8(approximate_lena), 'ca_lena.jpg');
% read approximate_lena layer in a variable
lena_ca = double(imread('ca_lena.jpg'));
% perform DCT on approximation level of lena picture
dct_lena = dct2(lena_ca);
% embed the watermark into dct_lena
% insert here function to embed the watermark
% dct_lena = insert_watermark();
% now I can recompose the lena coefficients approximation
lena_ca_recomposed = idct2(dct_lena);
% recompose DWT
lena_recomposed = idwt2(lena_ca_recomposed, horizontal_lena,vertical_lena, diagonal_lena, 'haar');
正如您在下面看到的：在第一行中，左侧是起始图片，右侧是最终图片（更暗，细节更少）。在左侧的第二行中，我们比较dwt2了原始图像上的 CA 和右侧重组后的 CAidct2
