Re: gzip vs. mtx compression ratios

Hi Thomas,

Any examples you would suggest?

Cheers,

John

----- Original Message -----
From: "Thomas Phinney" <tphinney@cal.berkeley.edu>
To: "John Daggett" <jdaggett@mozilla.com>
Cc: www-font@w3.org, "Philippe Le Hegaret" <plh@w3.org>, "chris" <chris@w3.org>, "Vladimir Levantovsky" <Vladimir.Levantovsky@monotypeimaging.com>
Sent: Tuesday, June 30, 2009 1:45:46 PM GMT +09:00 Japan
Subject: Re: gzip vs. mtx compression ratios

That's quite interesting. But for general fonts in the wild, neither
system fonts nor the ClearType C-fonts are very representative. They
are all hand-hinted, which makes them unlike 99% of TrueType fonts out
there.

It would be interesting to see the results on some typical TrueType
fonts that have been autohinted by (1) Fontographer 4.x and (2)
FontLab Studio 5. That would tell us a bit more.

Regards,

T

On Mon, Jun 29, 2009 at 2:05 AM, John Daggett<jdaggett@mozilla.com> wrote:
> I sat down a little today and played around with font compression.  I
> tested compression ratios of gzip vs. MTX, the MicroType Express
> algorithm used as part of the Microsoft EOT format.  The spreadsheet
> below contains charts and a summary of the data I collected:
>
> https://spreadsheets.google.com/ccc?key=rKT_wNzraVrkXQcKSWb-jTA&hl=en
>
> I used Microsoft's WEFT tool to create unsubsetted EOT versions of each
> font and compared it with the gzip-compressed version of the font file.
>
> As Vlad noted before, standard webfonts such as Times New Roman, Arial,
> Georgia and Verdana compress to file sizes 20-28% smaller than the size
> of files compressed with gzip.  However, fonts in the Cleartype font set
> that Microsoft commissioned for Vista are only 11-14% smaller.  These
> fonts use less hinting than traditional TrueType fonts and rely on
> Cleartype screen rendering to render glyphs clearly.  This is important
> because it seems indicative of a trend towards more lightly hinted
> TrueType fonts, which would indicate the hint-related compression that
> MTX provides won't be needed as much going forward.  But MTX for these
> fonts still does a better job than plain gzip compression.
>
> For large CJK fonts however, where compression is most needed, MTX
> doesn't provide much beyond straight gzip compression.  For Meiryo, EOT
> files are 17% smaller but for other CJK fonts the range was only 2-12%.
> In fact, using bzip2 general compression beat font-specific MTX for
> several of these fonts.  These numbers are also probably distorted in
> favor of MTX because the fonts were in TrueType collection files (.ttc)
> rather than straight .ttf files, so there are extra glyphs in the gzip
> file that aren't in the EOT version.
>
> Since the WEFT tool doesn't handle Postscript CFF fonts (.otf)
> currently, I tested the MTX compression of these fonts by merging CFF
> data from other .otf fonts into a TrueType font, then comparing the
> differences in the resulting compressed versions.  The MTX compression
> seemed to be around 5% better than straight gzip.  Although this doesn't
> account for compressed metrics, I think this is pretty close to the
> compression that a version of MTX modified to compress CFF glyph data
> would see.
>
> In summary, MTX seems to compress older TrueType fonts well but less so
> more modern lightly-hinted fonts.  It is a little bit better than gzip
> for CJK fonts but lags behind general bzip2 compression in many
> instances.  For .otf fonts it's only slightly better than gzip
> compression.
>
> Regards,
>
> John Daggett
> Mozilla Japan
>
>



-- 
"Men occasionally stumble over the truth, but most of them pick themselves up
and hurry off as if nothing ever happened."
- Sir Winston Churchill

Received on Tuesday, 30 June 2009 04:57:32 UTC