Does 'imresize' generate artificial pixel RGB values which are not contained in the original image?

Hi,

I would like to know whether 'imresize' generate artificial pixel values which are not contained in the original image.

For instance, if my image is 2 rows by 3 columns (extreme case, won't happen in the real life). There will be 6 pixels contained in the image.

For instance, if these 6 pixels are

[0 0 0], [1 1 1], [2 2 2], [3 3 3], [4 4 4], [5 5 5], [6 6 6]

After I imresize the image to 1/2. There will only be 3 pixels left. Will some of these 3 pixels be like [1.5 1.5 1.5] which does not belong to the original 6 pixels.

[Adam]

Yes.

If you don't want it to I'd suggest using

doc interp2

with 'nearest neighbour' interpolation method instead.

[Salad Box]

Thanks for your answer.

Answer 1

[Sai Bhargav Avula]

Hi,

Yes for interpolation methods like bilinear, bicubic. But if you want to avoid it.

As suggested by Adam you can try interp2.

I would also suggest something similar. The imresize has 'nearest' as it's Interpolation method. You can try that too.

    J = imresize(I, 0.5, 'nearest');

This shrinks by factor of two using nearest-neighbor interpolation. This is the fastest method, but it has the lowest quality.

[Salad Box]

Hi Thanks for your answer.

I don't fully understand about 'nearest neighbor' and why it has the lowest quality.

For instance, if I have 6 pixels in the order

[5 5 5], [0 0 0], [3 3 3], [3 3 3], [1 1 1]

How the nearest neighbor method work and why this is the lowest quality? Quality of what? How to improve the quality?

[Sai Bhargav Avula]

The major issue with this method is aliasing(jagged edges).

This is because it uses the nearest neighbour and applies it's intensity value.

To address this you can either use better interpolation method or Use spatial domain image filtering

You can refer the link below for further explanation https://www.mathworks.com/help/vision/ug/interpolation-methods.html