Re: [Devel] 'tension' in splines, Hermite Splines [Re: Rendering (by FreeType) and hinting of Latin small 'a']

[Vadim Plessky]

On Friday 13 September 2002 9:42 pm, Rogier van Dalen wrote:
|  Vadim Plessky wrote:
|  >| large amounts of delta hints, which are almost never needed when using
|  >| an anti-aliased rasteriser, in my experience.
|  >|
|  >|  Georgia, which is hinted far better than TNR, if you ask me, executes
|  >| only 99 to 103 instructions for "a".
|  >
|  >I like how you pronounce word "only" ;-)
|
|  Taking into account that  for "a" you do need to move at least 14
|  points, and the character is changed at low point sizes, this is not
|  much. It is not much more than you'd specify with PS hints or suchlike.

So, you think adding Tension/Continuity to node/curve will not help to get 
better rendering?
With TT auto-hinter in FontLab, 'a' is quite easy to hint.
I achieved quite good result using just about 18 "TrueType hints" which (as it 
seems to me) are more high-level hints than embedded TT OpCodes.

On the other hand, some glyphs (like Cyrillic Zhe Capital 'Ж' -U+0416, Zhe 
small 'ж'-  U+0436) need about 35 those *high-level TT hints* (in FontLab), 
and still rendering quality is low.

My wish is to release good-looking (professional quality) *free* fonts.
And I am not going to have release 5 years later. I want to do it in month or 
two.
Realistically speaking, I can manage to add/overview/edit about 2500 hints per 
font. (about 250 characters for Latin+Cyrillic alphabet, 10 hints per glyph)
Having 200 instructions per glyph is *out of question* for me.
I just can't add/check-out 50 000 (fifty thousands) instruction just for *one* 
typeface.  I just don't have so much time, and willingness to spend SO MUCH 
time, on hinting process... And you need 8 fonts in total, to make typical 
Desktop system usable...

BTW: that's what stopped me from working on free fonts for Linux 1.5- 2 years 
ago.
Than David added to FT PS hinter (about 1 year ago), which produced quite 
reasonable results, and this encouraged me to revive my work.
But *reasonable* doesn't equal to *acceptable*...
Finally, FT 2.0.9 had *acceptable* rendering quality for PS Type1 fonts, and 
this allowed me to work further.
In my opinion, we at 95% level of _being ready_ at a moment.
And I do not thinnk that we should throw away everything we have, and start 
rehinting all fonts in TTF (even supposing that such nice tool available and 
works fine, which is not the case at a moment)

On the other hand, it's quite interesting for me how other formats/hinting 
models/rasterizers solve problem I faced (like: rendering of 'a')
Here is experience I have:
* It seems to me that ATM 4.1 for Windows can't produce acceptable quality of 
rendering, either for PS Type1 or OpenType fonts
* same problem with rasterizer in Windows 2000.
While it seems to be betetr than ATM 4.1, quality of rendering is still low 
(pfb, otf)
* rendering quality of Times New Roman is consistant from Windows 98 to 
Windows 2000 (don't have WinXP and haven't tetsed, but should be fine, too)
But at waht price? Huge, HUGE TT programs! Several man-years of hinting 
work!..
* Garamond and Baskerville in Windows are not easy to recognize, and therefor 
not very useful. In best case, those fonts look like just as Times/Times New 
Roman. This probably explains the fact that Garamond, Baskerville, etc.are 
rarely used in office work (memos, documentattion, etc.)
And not that Garamond and Baskerville are *hinted* by MS, so I suppose quality 
of hinting in those fonts can be defined as *very good*...

Therefor, my conclusion (so far):
* TT hinting model is outdated and too complex.
There are no commercial applications/usage patterns justifying usage of TT 
hintng model
* PS hinting model is outdated. But it still has potential
* FreeType is the best renderer on the market, so far.
When used with AA, it really has no competition!..
* FreeType with AA enabled brings new *breath* into PS hinting model.
Note that you can, for example, distinguish Garamond from Times when they are 
rendered by FreeType (in contrast to Windows)

The only thing missing in current FT renderer is capability to protect glyphs' 
topology. I guess I already wrote about this before.
But as OCR programs exist, and most of them doing recognition via topology - 
topology *can* be added to raterizer.
Question is *when*, and *how*...
But it's better to have light at the end, instead of living in darkness!..

|  >|
|  >|  This is true, also because of the conversion from Bezier splines. I'm
|  >| not sure, however, this actually causes a lower font quality.
|  >
|  >Well, you will notice difference on professional typesetting machine
|  >(2500dpi), I guess.
|  >I also found printouts in Times New Roman difficult to read, as its
|  > outlines very artificially distorted to achieve best rendering results
|  > on screen.
|
|  O yes, of course. I thought you were talking about the hinting quality
|  decreasing because having more points tends to make instructing more
|  difficult. Presumably you should always use the original font for
|  high-res printing, which is a totally different matter.

Well, it seems I need to clarify my understanding of hinting/outlines/quality 
of rendering.
Peter Karow in his last book (1994) make a statement that "Times New Roman is 
the best Times from all variants (clones) of Times".

But note that he is refering to *original* Times New Roman (TNR) outline, 
which he was checking out in IKARUS format, I guess. He is refering not to 
TNR TTF!...

Microsoft certanly had a goal to produce easy-to-read (display) fonts for 
Windows (Win 3.1). They achieved that goal.
But ... at what price?
Price was huge. First of all, hinting of TNR required a lot of man-hours.
Second, outlines are distorted at almost all usable sizes (usable== used in 
modern office, not but Graphic Artists)
And, finally, TNR doesn't print very well on paper. I mean: it's not nice to 
read, especially on long texts.

My goal, eventually, to produce font both easy to read (on Linux/XFree86 
platform), and suitable for printing (both PDF and on paper).
None achieved both goals, on any mass-production platform.
But hey, that's why we are using OpenSource, right?!..
We can do much better than closed-sourced companies (MS, Adobe, Quark, Apple, 
Monotype, Linotype, etc.)
Life time of those companies is counted down. Linux/Open Source is growing.
In a few years, we will recall those companies as a joke, and wonder how 
people could use their programs few years ago... ;-))

|
|  Bye,
|  Rogier

Cheers,
-- 

Vadim Plessky
http://kde2.newmail.ru  (English)
33 Window Decorations and 6 Widget Styles for KDE
http://kde2.newmail.ru/kde_themes.html
KDE mini-Themes
http://kde2.newmail.ru/themes/