Another quick discussion related to font file format philosophy:
Is hinting really worth the effort?
Font file formats (even scalable ones) in principle ought to be
relatively simple creatures. The only aspect that really adds enormous
amounts of complexity, both with regard to the development of the
renderer as well as with regard to the creation of the fonts, is the
automated control point adjustment based on hinting information. Is that
type of scale-independent hinting really a good idea in the long run?
I'd like to argue that this is not necessarily the case, and would be
interested in hearing your more generic insights and opinions into the
subjects.
Main points:
- Even entry-level printing devices have now reached pixel sizes
of 20 µm or less, and many use in addition non-binary pixel
values for anti-aliasing, such that the changes of up to half a
pixel width to the outline when hinting is applied really does
not affect visible quality any more in printouts.
- For DTP applications, it is important that the on-screen
representation approximates as closely as possible the printed result
in a device-independent way, and hinting severely interferes
with that goal as it changes glyph spacings and sizes.
Experienced users know the problem that for example Microsoft Word
reformats a document if you change the resolution of the output device,
whereas Acroread uses unhinted glyph metrics in order to utilize
PDF as a truely device-independent format, which I find preferable.
- Neither web browsers nor GUIs have any need for exact freely
selectable glyph sizes. As long as a sufficiently smooth set of
glyph sizes with high-quality bitmaps is available, rounding to the
nearest font size that can be rendered in high quality is
perfectly acceptable (with say ~20% tolerance).
- The number of necessary bitmaps is relatively small. Glyphs less than
5 pixels high are unreadable anyway, with or without hinting, and
glyphs larger than around 50 pixels are sampled well enough such that
hinting starts to become unnecessary for a pleasant look, especially when
anti-aliasing is applied as well.
Therefore only for a single decade of glyph pixel sizes are manual
adjustments actually necessary. If the acceptable step size between
two font sizes is in the order of 20%, then manual adjustments are
necessary only for around 14-17 font sizes.
- Manually adjusting an unhinted glyph outline for a particular
low-res presentation can be done in one of two ways:
a) edit the resulting bi-level bitmap (switch some pixels
on or off to improve the looks)
b) manually snap control points to a nearby pixel boundary
Option a) is very easy to do and requires only readily available
bitmap software. I estimate that I can manually fix bitmap glyphs
at a rate of around 100 glyphs per hour. Option b) has the advantage
that it is compatible with anti-aliasing. I'm not sure how fast it
could be done; it would need a specialized software tool. The snapping
information can probabaly be compressed down to less than 4 bits
per control point and hinted scaling factor.
- I understand that the area of scale-independent hinting instructions
in font file formats has some patent problems. (?)
- In spite of the sophisticated hinting mechanisms available in
TrueType fonts, it seems that places like Microsoft Typography
do not make very significant use of it and embeds handedited
bitmaps instead.
So if we start one day some significant Free Outline Font project, shall
we really stick with existing TrueType/OpenType technology, or shouldn't
we skip the automatic hinting mess altogether and just add either
bitmaps or corrected control points for those scaling factors where,
let's say, the x-height is one of
$ perl -e 'for ($i=5; $i<=50; $i=int($i*1.2+0.5)) {print "$i, "};'
5, 6, 7, 8, 10, 12, 14, 17, 20, 24, 29, 35, 42, 50
or
$ perl -e 'for ($i=4; $i<=64; $i*=sqrt(sqrt(2))) {print (int($i+0.5) . ", ")};'
4, 5, 6, 7, 8, 10, 11, 13, 16, 19, 23, 27, 32, 38, 45, 54, 64,
The application would then specify to the renderer a font size and a
maximum tolerance (e.g., 20%), and if a hand hinted set of control
points is available within the tolerance, then the renderer will use
that that beautified on-screen font size instead of the exact requested
one.
(With regard to preferred numbers of font sizes, I believe that
neighboring fonts should be a factor 2^{1/4} = 1.1892 apart, because
this means for paper fonts that for example an A4->A3 magnification will
lead again to an exactly available font size. If you didn't understand
the last idea because you are North American, read then please have a
look at <http://www.cl.cam.ac.uk/~mgk25/iso-paper.html>.)
Just some food for thought ...
Markus
--
Markus G. Kuhn, Computer Laboratory, University of Cambridge, UK
Email: mkuhn at acm.org, WWW: <http://www.cl.cam.ac.uk/~mgk25/>
_______________________________________________
Fonts mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/fonts
