[CREATE] Blur Benchmarking: Synfig, Inkscape, Gimp, Anime Studio

[Jasper van de Gronde]

On 26-10-11 08:51, Carlos López González wrote:
>     > ...
>     One of the differences might be the color space used for the filter (and
>     blending), and this you can change in SVG using the color-interpolation
>     attributes. These are unfortunately not supported in Inkscape, but I
>     think most browsers do support it (not 100% sure).
> 
> 
> I don't want to cause a regression on any application. I'm just sharing
> with you those facts I've seen during blur comparison. In any case
> regressions can be always handled given the user to add one option to
> render as old version or as new version.
> 
> I'm not expert on the blur materia but intuitively I think that when you
> see something blurred you shouldn't get color deviation of it and only
> alpha channel should reflect that blur effect.

It's a little bit more complicated than that. Blur is defined by
specifying a kernel with which the image is to be convolved (per
channel). Inkscape does this, and as far as I can tell it uses the right
kernel (or at least a very good approximation to it). From a technical
standpoint this does NOT result in any "color deviation" or whatever in
terms of the color values (you can check this yourself by looking at the
intermediate colors, you'll see that they nicely interpolate the two
extremes).

However, you, as a human, might still perceive a color deviation. This
can have several causes, not the least of which is the color space in
which the blur is performed (imagine doing it in HSL space, now that
would be "interesting"). SVG provides two options for authors, sRGB and
linearRGB. Have you tried using color-interpolation-filters=linearRGB?
It's currently not supported by Inkscape (it's the default for filters,
but not for other operations...), but might be supported by some of the
browsers.

So basically what I'm saying is that what you're seeing is not a bug,
it's a feature (and there are some good reasons for that). And if for
some reason you don't like the result you should try using different
methods to obtain a result that better suits your needs.

> ...
> Similar happens to the antialiasing effect. See this antialiasing
> comparison [1]
> <https://docs.google.com/document/d/17pNfb7TKw_wzAt6fGcDzQ5L5-Ms8iQka1VrLwMnKkNw/edit>.
> It shows how Inkscape and Anime Studio create color biased pixels when
> exporting a single color shape to png. This will cause antialiasing
> problems when further processed.
> 
> On the example, Gimp and Synfig does exaclty the same job but Inkscape
> and Anime Studio produce color biased pixels. I'm not aware on what
> other applications do and might be interesting to know it.

I just tried to recreate the result (using GwenView, as you didn't give
the ImageMagick commands), and as far as I can tell Inkscape's output is
just fine. Have you tried using the output in any other context? (Like
in a browser, or using an image viewer.) If you're still having
problems, it would help to know a bit more details, so we can determine
where the problem lies (I would suggest doing this through a bug report,
see link below).

BTW, we do have Inkscape-specific mailing lists (one for users and one
for developers). And if you want to file a bug, see:
http://bugs.launchpad.net/inkscape/