Re: Optimising output for screen.

[Kaz Kylheku]

On Fri, 24 Sep 2010 15:49:15 +0100, "Phil Burfitt"
<address@hidden> wrote:
> Trevor, I've downloaded a number of PDF viewers for Windows and taken
> screen snapshots at 96dpi on Vista of each at 100% scaling (PNG
> attachment). Problems occur in all of them as to be expected as
> Lilyponds output is optimised for printer and will have a resolution
> of probably at least 300dpi.

In fact, Lilypond's output doesn't have a resolution. It is vector
graphics. The music symbols you are seeing are drawn using Bezier
curves and lines.

It is up to the renderer to optimize this.

 The most common problem seen in these
> viewers is the rendering of staves that when scaled down do not fall
> exactly on a pixel. Adobe doesn't seem to anti-alias raster images,
> but it's font rendering is a lot sharper. Evince does the best overall
> job IMO.

I have both installed and they are producing
roughly equivalent output. Adobe is obviusly anti-aliasing
the graphics. However, it does a terrible job when the size is reduced
(magnification < 100%). Adobe at 50% looks like crap compared
to Evince at 50%. Beams on sixteenth notes are glued together
in the Adobe output, but in the Evince output, they look
perfectly distinct. At the larger sizes, like 100% and up,
it's fine. It seems to have anti-aliasing in the curve rendering
algorithm itself, but fails to implement a decent filter when
scaling down. To do that right you have to keep rendering the
graphics at a higher resolution and do a shrink; you can't
continue to rely on the anti-aliasing in the rasterization
routines for curves and lines. Adobe seems to do a very crude
supersampling job when shrinking. It's not as bad as one sample
per pixel, but maybe as little as four.

Adobe Reader's magnification goes up to 6400%. You can use
this to see that the graphics really is resolution independent.
A treble clef viewed at 6400% appears perfectly smooth;
there is no evidence that pixels are being magnified.

For reference, I have here Adobe Reader 9.3.3 and Evince 2.30.3.

I do see the bar lines over-extending past the staff lines.
This must be a Lilypond bug.

If that were a scaling artifact, it would be impossible to
draw the letter T without the stem crossing through the
crown.