Digital Watermarking - Science topic

% Clear the existing workspace

clear all;

% Clear the command window

clc;

% Read the input image

input = imread('peppers.png');

% Convert image to greyscale

input=rgb2gray(input);

% Resize the image to required size

input=imresize(input, [512 512]);

% Message to be embedded

message='geeksforgeeks';

% Length of the message where each character is 8 bits

len = length(message) * 8;

% Get all the ASCII values of the characters of the message

ascii_value = uint8(message);

% Convert the decimal values to binary

bin_message = transpose(dec2bin(ascii_value, 8));

% Get all the binary digits in separate row

bin_message = bin_message(:);

% Length of the binary message

N = length(bin_message);

% Converting the char array to numeric array

bin_num_message=str2num(bin_message);

% Initialize output as input

output = input;

% Get height and width for traversing through the image

height = size(input, 1);

width = size(input, 2);

% Counter for number of embedded bits

embed_counter = 1;

% Traverse through the image

for i = 1 : height

for j = 1 : width

% If more bits are remaining to embed

if(embed_counter <= len)

% Finding the Least Significant Bit of the current pixel

LSB = mod(double(input(i, j)), 2);

% Find whether the bit is same or needs to change

temp = double(xor(LSB, bin_num_message(embed_counter)));

% Updating the output to input + temp

output(i, j) = input(i, j)+temp;

% Increment the embed counter

embed_counter = embed_counter+1;

end

end

end

% Write both the input and output images to local storage

% Mention the path to a folder here.

imwrite(input, 'path_to_folder\originalImage.png');

imwrite(output, 'path_to_folder\stegoImage.png');

If you need Pixel adjustment code which provides better PSNR than simple LSB then contact with me.

Dear all

I need a base paper to implement my algorithm in hardware. if anyone working in this area kindly share those details

I think so.

Search on www.Mathworks.com, Many codes are available there

link for DCT watermarking and pso matlab code

Cryptography: change the data so it is not readable. Adversary can see there is a data communicated but can’t understand it.

Steganography: hide the very existence of the data. Adversary doesn’t know of a secret communication.

Watermarking: either visible or invisible and used to identify ownership and copyright.

you can use PSNR and structural similarity index metric (SSIM) to evaluate your algorithm.

Yes, it's easy to program your watermarking algorithms in OpenCV especially with C++. But you must code them because in my knowing it doesn't exist yet a predefined openCV function for this purpose.

There is a solution: we can get an uncompressed video from frames using this command

ffmpeg -i ./frame%d.png -vcodec png video.avi

In a program, the consistency constraints with shared data structures.are called Colors.   Generally a programmer will be assigning colors to data. As a part of coloring,  the programmer also marks places in the control flow where data should be made consistent.

Water mark is computed for a given digital object and placed along with the same object. Water mark is basically digitally signed emblem. This can be verified by any third party similar to that of process used for verifying the digitally signed certificates.

I think you can follow this discussion. 

Please go to the definition of the effective payload given between Eq. (1) & Eq. (2) in the second page of the paper.

Thank you mr Ersin

Payload depends upon the data you want to enter but with increasing the payload, your quality of the watermarked media file will going to be reduced. There must be a trade off between the payload and quality. So, this trade off depends upon your application. 

In basic QIM, a quantizer maps a value to the nearest point belonging to a class of pre-defined discontinuous points. The function round (.) here denotes rounding value to the nearest integer and the standard quantization operation with step size is defined

as QΔ(x) = round(x/Δ)Δ.

Let Δ be the quantization step size and L represent the length of the host signal x and the message m (we embed one bit per sample). QIM embeds a message by modulating an index or sequence of indices with a watermark to be embedded and then quantizing the host signal with the associated quantizer. For QIM with dithering, we choose d0 pseudo-randomly with a uniform distribution over [-Δ/2, Δ/2] and

d1 = d0 + Δ/2,   if d0 <0

otherwise

d1 = d0 - Δ/2, if d0 >0

Here d0 or d1 is used for embedding message bit “0” or “1” respectively. The watermarked signal is given by:

yn = QΔ( xn , dm)

       = QΔ( xn + dm) − dm                             m∈[0, 1]                 n=1, 2, …, L

where QΔ(x) = round(x/Δ)Δ and the function round (:) denotes rounding value to the

nearest integer.

At the detector the received signal z, possibly a corrupted version of y, is requantized with the family of quantizers used while embedding to determine the embedded message bit, i.e.

As eqn are difficult to write refer my paper

or

thank you so much  Maddali V S Murali Krishna for your answer. in,fact I want to know practical tutorial or book or website that explain to me. how to implement DEW practically  with video (step by step) even in c programming language, c++  or c#

please if you or anyone can help me

The answer depends whether the product you are testing the legitimacy of is either software or printed matter. For software text documents digital signature is widely used and there are many models for you to try.

I agree with Peter. Adding a pseudo-random sequence in the time domain is equivalent to adding a pseudo-random sequence in the frequency domain because the transformation is usually linear and invertible. So one choose one domain rather than another because:

- one has a good perceptual model in this domain enabling to weight the watermark signal energy.

- one expects some attacks which are better modeled in this domain so that their impact on the watermark is better understood.

I strongly disagree with people making a clear difference between time and frequency watermarking, and claiming that frequency domain is better.   

This Matlab code might help you to understand about reversible data hiding or reversible watermarking based on prediction error expansion and histogram shifting.

I have been studying  reversible data hiding or reversible watermarking for many years. I think reversible data hiding based on prediction error expansion and histogram shifting is the best  technique so far.

If the Matlab code has some mistakes, please let me know.

Good luck.

 

  

This model takes into account some features of Human Visual System. A nice describtion of Watson model can be found in

I.J. Cox et al ''Digital Watermarking and Steganography''

Thank you Aria

I Think you don't pay attention to the matter that the image representation in spatial (color) domain contains only integer numbers, so if your watermarking is sensitive to the fractional part it will be discarded.

At the most of the practical methods, parameters are set in way that the method became robust against this quantization error.

Check the following, it may be useful to you,

1. Format of the image (imread, imwrite).

2. Transform and Inverse Transform must be applied correctly (levels).

3. There must be a solid reason to apply SVD.

Frankly I am unaware of any watermarking optimisation method for a completely unknown attack. Your question is more related to signal processing attacks. You can possibly generalise the attack characteristics and investigate the effected image regions (mostly high frequency components). However if the attack is intentional, there are few works available in the literature, e.g. 'Oracle Attacks and Covert Channels' by Ilaria Venturini. This may lead you to create a similar model for signal processing attacks.

Additionally, can any one inform me the references w.r.t such a work that I can access?