0
\$\begingroup\$

If a scenic photograph is scaled double its original size to conform to some printing dpi/inches criteria, the default setting in Gimp appears as "cubic interpolation".

Assuming this is the best option when scaling images, does scaling deteriorate the quality of the overall image in some way?

Pixelation is very worrisome, but would like to know more detailed info by how much it might occur, or other considerations I should know of when scaling images.

\$\endgroup\$
5
  • 1
    \$\begingroup\$ When you say "Scale the image larger" are you talking about increasing the pixel count while keeping the same display size? or increasing the display size along with the pixel count? \$\endgroup\$
    – Michael C
    Commented Dec 26, 2021 at 5:29
  • \$\begingroup\$ Higher pixel size. Display size is meaningless if you actually mean aspect ratio (this is kept intact when rescaling pixels) \$\endgroup\$
    – user610620
    Commented Dec 26, 2021 at 8:29
  • 1
    \$\begingroup\$ Display size is not meaningless. It is a part of the equation as to how many pixels you need before you can see individual ones. This also applies to print. There are some pretty smart upscalers these days, but none of them can create information that wasn't in the original. \$\endgroup\$
    – Tetsujin
    Commented Dec 26, 2021 at 9:01
  • 1
    \$\begingroup\$ @user610620 "Display size" means exactly that. The physical dimensions at which the image is displayed and viewed. Image files have aspect ratios, but they do not have physical dimensions. Displayed images have physical dimensions. \$\endgroup\$
    – Michael C
    Commented Dec 26, 2021 at 10:58
  • \$\begingroup\$ ok so the question is about pixel size not corresponding to display size (inches are physical dimensions, and therefore absolute display sizes) \$\endgroup\$
    – user610620
    Commented Dec 26, 2021 at 12:32

2 Answers 2

1
\$\begingroup\$

If you upscale the picture you don't get pixellation (unless you use the "None" interpolation (technically the "nearest neighbor")) you get something blurry. In modern Gimp (2.10) the best interpolations are "NoHalo" and "LoHalo".

The image quality come from the amount of information in it, which is related to the number of pixels. But when you upscale the picture you are not adding any information you are just spreading it on a greater area, hence the blur.

Upscaling is likely not necessary, see this.

\$\endgroup\$
2
  • \$\begingroup\$ Pixellation and blurring seem very related. How to differentiate them? What's so bad about quadratic interpolation compared to NoHalo and LoHalo? \$\endgroup\$
    – user610620
    Commented Dec 25, 2021 at 18:30
  • 2
    \$\begingroup\$ Pixellation is very different from blur. Cubic (not quadratic) is not awful (it used be be a good default because results are often OK without using too much CPU), but NoHalo/LoHalo are better (less susceptible to aliasing IIRC). \$\endgroup\$
    – xenoid
    Commented Dec 25, 2021 at 18:49
0
\$\begingroup\$

The question is a bit broad, and it is a bit complex to answer. In the end, it is probably more a thing about expectations vs reality.

Every resampling algorithm tries to guess what a new pixel between two existing ones should look like. In some cases, they average the values around, so the result is hiding the pixels behind a blur.

Sometimes that is just what you need. Not having distracting pixels. If the image is interesting, this slight blur will not distract you as much.

Newer AI algorithms try to make better guessing by analyzing the context and the relationships between pixels, not only the adjacent ones, but along some patterns, for example, they try to see if the pixels are from a sharp contrast zone.


In my opinion, I only would resample an image at exactly 200%, and only to hide an already small pixel. For example, in most cases, I consider a print of 100PPI good enough, even to be viewed close, let us say 30-50cm. But to avoid seeing the pixels on sharp edges, I double it, having a final resolution of 200PPi.


The point is not trying to fool yourself into thinking that you achieve great sharpness only by resampling an image.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.