Upscale and downscale = better quality?

[Quantumleap]

Hello,

I went through a few posts as I wanted to archive some of my old cartoons.

Now I've been reading that to get a better quality an upscale and then a downscale should be performed.

I do not understand how this improves quality.

Can someone please explain what is the logic behind this process?

Thank you in advance.

[pandy]

Generally you can process pixels in more refined way - for example you can increase pixel 4 times using Nearest-neighbor filter then apply some filtering (like Median) and finally downscale product with Lanczos - by increasing image, processing, downscale you can get results in sub-pixel space.

[LigH.de]

I guess the upscaling is to be done with an EDI (Edge Directed Interpolation) filter, which is tuned to enhance straight lines in a sharp style without oversharpening (ringing artifacts). Good for cartoons, this will enhance the line drawing appearance and keep the flat color areas between the lines smooth and flat. Applying the same to general video content may not always look just as great...

Using AviSynth to process the video, you may see the NNEDI3 plugin being used with the function nnedi3_rpow2 which can be used to upscale video by powers of 2 (2, 4, 8, 16 ... times the original width and height).

[Quantumleap]

Ok.
One thing I don't get is, does this work in one task?
Meaning, do I have to upscale and save the file and then downscale the new one?
Or all in one process?
I'm relatively new to Avisynth scripting.
Sorry, for all the questions that might sound stupid.

[Selur]

If you upscale&filter&downscale-combos ifyou know that the filtering you are doing in the middle:
a. really profits from it
b. still work as intended (for example deblocking filter might not properly detect blocking)
c. don't themselves also increase and decrease the resolution to do "super sampling"
-> the assumption that it's always a quality improving step to upscale and later downscale is wrong (especially if the up-/down-scaling additionally introduces artifacts)

That said: the whole thing is usually done in one script. (quite a bit of filter scripts already do it themselves)

My recommendation would be: if you are new to Avisynth and video filtering in general do not mess with this.
If you know your video/image processing algorithms and convolution, fourier transform and general signal processing are nothing 'new' or 'unknown' to you and you know what a filter does it might be helpful.

Cu Selur

[Quantumleap]

If you upscale&filter&downscale-combos ifyou know that the filtering you are doing in the middle:
a. really profits from it
b. still work as intended (for example deblocking filter might not properly detect blocking)
c. don't themselves also increase and decrease the resolution to do "super sampling"
-> the assumption that it's always a quality improving step to upscale and later downscale is wrong (especially if the up-/down-scaling additionally introduces artifacts)

That said: the whole thing is usually done in one script. (quite a bit of filter scripts already do it themselves)

My recommendation would be: if you are new to Avisynth and video filtering in general do not mess with this.
If you know your video/image processing algorithms and convolution, fourier transform and general signal processing are nothing 'new' or 'unknown' to you and you know what a filter does it might be helpful.

Cu Selur

Hello Selur,

I don't expect wonders from upscaling.
I mean I know that I can't get what's not there.
But I would like to learn, that's why I'm here.
I appreciate your feedback and will have a deeper look into the available filters that are out there.

Thank you.