[cairo] LCD filtering in cairo, use default filter?

[Carl Worth]

On Sun, 13 Jan 2008 16:41:38 +0100, Sylvain Pasche wrote:
> I've setup a page comparing the various LCD filtering with a set of font
> at different sizes, either with the bytecode interpreter or the
> autohinter (warning, large page):
>
> http://spasche.net/files/lcdfiltering/

Wow, thanks. This is really nice to see.

I'd seen lots of requests to change cairo's behavior to "reduce color
fringing" but no head-to-head comparison that really showed the full
matrix, (particularly with and without the bytecode interpreter),
until this. So thank you! This is invaluable.

> As expected, when using the bytecode interpreter the type of LCD
> filtering has not much impact on the rendered images. The glyphs with
> the legacy filter are a bit thinner that the one with the default one.

I disagree on "not much impact" here. To my eyes, I see a big
cyan-tinted blur around the glyphs with the DEFAULT and LIGHT filters
compared to LEGACY. (This is with Sans 6 and Sans 8 with the bytecode
interpreter on).

> When using the autohinter (which is the default on some distributions
> like Fedora because of patent issues), the difference is more
> noticeable. The none filter (first column) shows visible color artifacts
> which are not very pleasant. While this is less visible, I can see more
> color fringes on my screen with the legacy filter in comparison to the
> default one.

I still see lot's of cyan blur with DEFAULT and LIGHT. A word with
lots of vertical elements, ("brillig" is a great one for example),
really shows off how bad it can be. I will grant that the LEGACY
color fringing on a glyph like 'w' is quite objectionable. (Again,
this is with Sans 6 and Sans 8, but this time with the autohinter.)

> Before going further, I'm wondering if it makes sense to make this value
> configurable or if we should just always use the default filter. When
> using the bytecode interpreter there is not much difference and it can
> improve the output when using the autohinter.

If that's your conclusion, then we have to have configurability I
think. I see a general regression in readability switching from
LEGACY to DEFAULT, (and this is definitely a bigger regression with
the bytecode interpreter on).

So I think if we're going to do any change here we'll have to chalk
our different interpretations up to personal preference and make
things configurable to try to make everyone happy.

-Carl
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.cairographics.org/archives/cairo/attachments/20080114/355d92c0/attachment.pgp