Re: [Devel] Dropouts in AA text

[David Turner]

Hi Keith,

Keith Packard a écrit :
> 
> Using 'timesi.ttf' (Times New Roman from Microsoft's webfonts page), I'm
> losing the diagonal element in the lowercase 'z' at sizes from about 8
> pixels up to about 22 pixels.  At larger sizes, the diagonal is vaguely
> visible, at smaller sizes the vertices are not snapped to a grid and the
> diagonal looks correct.
> 
> When rasterized to a monochrome bitmap, the diagonal is visible down to
> sizes around 9 pixels; after that it's just a mash of pixels.
> 
>         http://keithp.com/~keithp/download/dropouts.png
> 
> shows the problem.  Note that the displayed numbers are point sizes, my
> monitor is 86dpi making the pixel sizes 1.2 times the point sizes.
>

The problem you're experimenting comes from the fact that the two diagonal
edges of the "z" are simply superposed, resulting in a covered pixel area
of exactly 0. It is "normal" for vector graphics not to display anything
in this case, though sometimes judged un-esthetical by users. You could
see the same defects with the monochrome scan-converter if it didn't include
drop-out control algorithms.

It seems you're using the TrueType interpreter for the hinting. Most TrueType
glyphs are hinted for the monochrome renderers of the Mac and Windows, and this
sometimes gives surprising results when using the FT2 anti-aliaser.

Another example would be the capital "D" of Georgia at certain sizes, whose
ascent and descent seem too large at certain sizes _because_ of its hinting..

There is no generic way of solving the problem of "incorrectly" hinted
glyphs. We could try to provide drop-out control in the anti-aliasing
renderer, but this would require significant work. Another solution is
to use a monochrome renderer to generate 4x times bigger glyph _with_
drop out control, then filter them down..

Note that the BeOS and QNX Photon, which both use a TrueType interpreter,
exhibit the same "missing bits" when anti-aliasing many fonts..

Regards,

- David