Can I undo Interp2 function?

I use code bellow to resample my Image

for i=1:z

      nimg(:,:,i) = interp2(Ox,Oy,img(:,:,i),Nx,Ny,intmethod);

end

Can I undo interpolating to receive first Image?

Answer 1

[Matt J]

Approximately, yes, depending somewhat on what the transformation is and whether it is invertible. For example imrotate(Img, theta) does a specialized form of the resampling you've shown. The inverse rotation imrotate(img,-theta) would return an approximation of your original image.

[bkshn]

How can I find that my transformation is invertible? I use interp2, is it invertible?

[Matt J]

The specific transformation you are performing with interp2 has to be analyzed for invertibility on mathematical grounds. In general, if the Jacobian determinant of your transformation is non-zero through the image, the transformation is invertible.

Common kinds of invertible transformations are rotations, translations, certain affine transformations, etc... See documentation for tformfwd() and tforminv() for additional common kinds of invertible transformations.

Instead of interp2, you might even consider using maketform(), imtransform() and tforminv(), since they make transform inversion simpler.

Answer 2

[Walter Roberson]

Take a 20 KHz sine wave. Resample it 5000 times per second. Lots of resolution! Now resample those 5000 samples per second back to 20000 KHz. Plot it out, and you'll see a good approximation of the original signal. Oh, wait, the plot is flat ?!? Better resample it 10000 times per second for better accuracy ?!?

[bkshn]

I want to resample an image not a Wave, I don't know is it your solution good for me?

Answer 3

[Matt J]

If you were hoping to invert not just the continuous space transform approximated by interp2, but in fact the whole discrete image transformation it performs, you could try FUNC2MAT.

Basically, you can convert the transformation W=interp2(V,XI,YI) to a matrix equation

   W(:)=A*V(:)

where the matrix A is obtained by

    A=func2mat(@(z)interp2(z,XI,YI), zeros(size(V)));

You can then attempt linear algebraic pseudo-inversion

      V_recovered=reshape(A\W(:), size(V))

However, the invertability of A would depend greatly on XI, YI and in an unclear way.