[MPlayer-users] Comparison of different software scaler types

[D Richard Felker III]

On Tue, Oct 14, 2003 at 04:29:22PM +0200, Matthias Wieser wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> Hi,
> 
> I have done a small webpage, showing the effects of different software 
> scaler types (-sws x).
> 
> First I used mplayer to scale video frames down to 50%. Those frames were 
> saved as PNGs. Then I used imagemagic to scale them back to original 
> size, so you don't need a magnifying glass to see small details.
> 
> At http://www.wieser-web.de/MPlayer/ you can see the differences.
> 
> Results:
> - sws 0    produces acceptable results
> - sws 1    produces blurred images
> - sws 2    produces very good images
> - sws 7    is even worse than -sws 1
> - sws 8    produces shadows near strong edges and increases contrast
> - sws 9/10 produce the fewest artifacts and gives even a bit better
>            results than -sws 2

I looked at your pictures, and IMO you misunderstand what these are.
There is no correct (much less no unique correct) way to scale images
down. You have lots of different choices, and each sacrifices certain
things. Of course a Gaussian process (7) is going to blur an image! At
first glance it appears to be a bug that 1 blurs the image while 0
doesn't (they're both bilinear), but in fact 0 is a hacked/optimized
version that only does correct bilinear scaling for upscaling.

If you want a better test, you should use MPlayer to scale the images
*UP* with different filters, and then scale them back down and compare
to the original.

Rich