matlab - steganography using DFT -

i have problem getting image after hiding it, converted hidden image to
dct cannot back, cannot find problem in code:

   a=imread('e:\sofia1.jpg');     m=imread('e:\mandrill.jpg');     [ax b]=size(a);    [cl d]=size(m);     a=imresize(a,[250 250]);    m=imresize(m,[250 250]);    a=a(:,:,1);    m=m(:,:,1);    mf=fft2(m);       mfbw=im2bw(real(mf));     steg=a+uint8((mfbw));  %hiding  image mfbw  in image        m=bitget(steg,1);   %retrieving data      h=fspecial('gaussian');      [c,d]=size(h);     hp=freqz2(h,size(mfbw,2),size(mfbw,1));        hps=ifftshift(hp);     g=hps.*mfbw;   % product of transfer function , image in frequency domain      g=real(ifft2(g));     imshow(g)      i=mat2gray(g);     imshow(i) 

if want use dct, may dct2, instead of taking real part fft2, because it's faster , more straightforward.

regarding question, crucial problem.

mfbw=im2bw(real(mf)); 

this creates binary image of dct coefficients. means values either 0 or 1. a binary image of dct coefficients has pretty no relation original image, m, looks like.

a less major, still important flaw in following line.

steg=a+uint8((mfbw));  %hiding  image mfbw  in image 

you add these binary values cover image. example, if least significant bit (lsb) of pixel in 1 , add 1 mfbw, bit change 0 (with carry flag next bit). want substitute lsb of pixels in 0s , 1s. done following command:

steg=bitset(a,1,mfbw); 

but remember, information embed nonsense.

final thought: seem misunderstand steganography in frequency domain means. transform cover image frequency domain, hide secret there , convert spatial domain. dct specifically, cover image split in 8x8 blocks, transformed individually.