please try this code

function [ cA, cH, cV, cD ] = myDWT( matrix )

%MYDWT Summary of this function goes here

% Detailed explanation goes here

matrixHeight = size(matrix, 1);

matrixWidth = size(matrix, 2);

cA = zeros(matrixHeight / 2, matrixWidth / 2);

cH = zeros(matrixHeight / 2, matrixWidth / 2);

cV = zeros(matrixHeight / 2, matrixWidth / 2);

cD = zeros(matrixHeight / 2, matrixWidth / 2);

for i = 1 : 2 : matrixHeight

for j = 1 : 2 : matrixWidth

w = matrix(i,j);

x = matrix(i,j+1);

y = matrix(i+1,j);

z = matrix(i+1,j+1);

a = ceil(i/2);

b = ceil(j/2);

cA(a,b) = floor( (floor((w+x)/2) + floor((y+z)/2)) / 2 );

cH(a,b) = floor( (w-x+y-z)/2 );

cV(a,b) = floor( (w+x)/2) - floor((y+z)/2 );

cD(a,b) = w-x-y+z;

end

end

please refer the following link also

http://stackoverflow.com/questions/25043858/2d-dwt-steganography-using-convolutional-codes

Most of the researchers use LL band for embedding secret information in case of steganography.  Even in watermarkin also, people choose LL band for embedding watermark. The reason is straight forward. Out of all the four bands, LL band has maximum information about the image. Some have tried insertion in all the bands also.  The ideal choice in my opinion is LL + LH / HL.  

Since insertion is done only in the HH band, which mostly contain high frequency coefficients, the retrieved information is noisy. 

@Chandra Mouli P.V.S.S.R - In LL sub band insertion the distortion is much high and retrieval is not possible even from TIFF image! Actually, I feel the problem lies with my bit selection..I am trying to replace an entire bit plane with secret message instead of selective insertion and may be there is the problem. I have to first select the bits those will provide imperceptibility and better retrieval but don't know which bits to consider!!

Anybody have soft copy of the book "Digital Watermarking and Steganography" by cox et. al.?

What do you need exactly?

I am trying to implement the transform domain stego algos (DWT/ DCT based methods) and study the properties of those. But after implementing the simple  DWT model I discovered that I am unable to retrieve the  secret the way I used to get in the spatial domain implementations ! Transformed domain methods are supposed to be robust against various statistical attacks but in contrary I am losing my secret, may be due to some basic mistake in my understanding. So, I want reference of some good papers or a good book that discusses the transform domain stego model.

Try to check the paper titles

1. "Performance evaluation of DWT based image steganography"

DOI: 10.1109/IADCC.2010.5423005

2. "Robust image steganography technique based on redundant discrete wavelet transform" DOI: 10.1109/ICPCES.2012.6508088

It will help you to find the spatial and frequency domain methods in stegangraphy approach

To retrieve non-noisy secret, you have to ignore HH sub-band. Better to use LL.

Good luck.

Thank u all !!

@Minati Mishra have you get the solution of your problem?? actually i am also trying to hide some info using 2DWT on image ,i am replacing some coefficients of HH band but on retrieval values of secret message have changed. can you please guide me??

thank you

@Neha Gautam

Yeah, i have got the solution. will post it for u..plz allow me a little time..presently extremely busy with academic assignments. If it's not very urgent for u, i would love to discuss the topic after a while

Try to use integer wavelet transform instead of DWT as IDWT is non-lossy. I hope this will help you to retrieve the entire embedded message. 

@Minati Mishra

ok,whenever  u will free plz guide me.

thank u.

@ Saiful Islam 

I have tried integer wavlet transform but again it leads to same problem

thanks for your suggestion.

@ Minati Mishra as  you mentioned in your earlier post that you will post the solution for this when you will free.if you have time so kindly post the solution i need it.

thank you

I was trying the DWT (a variant of HAAR DWT?) and the psnr is coming to be as low as 10.69 simply after the single level decomposition and recomposition steps! The further decomposition and payload embedding certainly will reduce the psnr further. Anybody worked out this scheme?

@Priyadarsini: you r right! I too encountered the same problem. Even a little embeding into the LL subband results with sever loss in image quality. 

If you solve the problem plz tell me