Problem with imresize function.

The problem i am facing is that imresize is not a reversible function.For example,if you convert a binary array of size 5x8 to an array of size 1x192 and then convert it back to size of 5x8,the values are not the same.In case,if u use 'bilinear' attribute then it gives back binary values,however,the values don't match.Please guide me how i can overcome the above problem.(Note:its not essential that the above problem be solved using imresize function.)Any help would be appreciated.

Answer 1

[Image Analyst]

You need to use reshape, to get it linear so that you don't lose any elements, and then use the 'nearest option of imresize. Check out this demo:

% Generate sample binary data.
m = randi(2, [5 8])-1

% Now do what Anish wants to do.
mLinear = reshape(m, [1 numel(m)])
% Now make it 192 elements long, for some reason.
mLinear192 = imresize(mLinear, [1 192], 'nearest')

% Undo the process:
mLinear2 = imresize(mLinear192, [1 numel(m)], 'nearest')
m2 = reshape(mLinear2, [5 8]);

% Subtract to see if we did it correctly;
difference1 = mLinear2 - mLinear  % Check stage 1.
difference2 = m2 - m % Check final stage

Why do you want to do this anyway? What does this do that you can't do when it's left in its 2D form?

[Anish Surana]

Thanx a lot...it works the way i want it to be.I need it for my steganography project(DWT with HVS) where i am supposed to have a row vector which is later to be converted into three matrices for embedding purpose...

Answer 2

[Jan]

Resizing an image until it has a single row only cannot be reversible. Reshaping seems to be more useful, if the problem is not restricted to imresize:

x = rand(5, 8);
y = zeros(1, 192);
y(1:numel(x)) = x(:)';

And backwards:

x2 = reshape(y(1:(5 * 8)), 5, 8);

[Anish Surana]

Thanx for the answer.It works well.However,having these many zeroes in y would be too redundant for my purpose.

[Jan]

Then, of course, you can omit the zeros and use this:

y = x(:).'

Answer 3

[Geoff]

The upscale and subsequent downscale of 8 > 192 > 8 should be fine, but I don't know how you expect resizing from 5 > 1 > 5 is going to give you back the same information. When you collapse 5 lines down to 1, you lose all the information in those lines. You cannot scale it back up to 5 and expect your original data to come back out.

Think of this:

R = rand(5,1);
M = mean(R);

How do you recover R using only M?

[Anish Surana]

Thanx for the answer...what i was concerned was the elements(in the same specific order like a bit stream) and not the rows and columns.