Re: meeting minutes 4/4/2013

On Fri, Apr 5, 2013 at 1:36 AM, Jasper van de Gronde <
th.v.d.gronde@hccnet.nl> wrote:

> On 05-04-13 00:32, Dirk Schulze wrote:
> > ...
> >    <scribe> ACTION: Tav produce an exmaple, demonstrating how
> >    different color space interpolation effects image scaling
> >    [recorded in
> >    [13]http://www.w3.org/2013/04/04-svg-minutes.html#action02]
> >
> >    <trackbot> Created ACTION-3481 - Produce an exmaple,
> >    demonstrating how different color space interpolation effects
> >    image scaling [on Tavmjong Bah - due 2013-04-11].
>
> In case you're interested, this has a nice discussion of the effect of
> gamma on resizing images:
> http://www.4p8.com/eric.brasseur/gamma.html
> As far as I can tell this is the main issue with the bitmap part, and
> I'd say it's excusable, given that SVG currently at least seems to
> suggest doing this kind of operation in sRGB. (Although this might
> indeed be something to work on.)
>

Yes,
the color becomes different because of gamma. When up/downscaling, white
and black are average.
However, 50% gray don't have 50% luminosity so the result looks darker.


>
> The issue with the top-right square is quite interesting. My guess is
> that it's related to the "seams" problem that you see when you two not
> quite pixel-aligned boundaries right next to each other, in that this
> problem probably also stems from an imperfect estimation of the coverage
> values because of not being able to determine whether different parts of
> the image overlap or not. I'd love to hear what you find, as I cannot
> quite put my finger on how exactly the observed value is arrived at.
> (Without basically retracing the steps of the rasterization algorithm.)
>

That's related how scan conversion is done. How much coverage do you need
in a pixel before it's covered?
For instance, if you draw 1 pixel lines in Illustrator and put them 1 pixel
apart, you just get a black square when you look at 100% zoom (where 1px =
1 device pixel) unless you're really careful where you place it.
This seems to be what IE does too as it just show a black square.

Surprisingly, WebKit and Firefox align the pixel exactly with the screen.
If you drag the window you can see the flashing to black before they
recompute.
I also tried on a chrome pixel. That one gets it wrong by aliasing the
vector art.

It seems that scan conversion and pixel snapping should be specified
somewhere...