Morphic 3.0: The future of the Gui
Joshua Gargus
schwa at fastmail.us
Thu Aug 30 16:27:19 UTC 2007
I haven't read the paper, but the video explains the idea quite
well. The algorithm operates at the pixel level. When shrinking, it
removes pixels without adjusting the newly-adjacent pixels, and when
growing it creates pixels by using the average of the pixels that it
is to be inserted between. I think that this technique is great,
because it provides good results, yet is MUCH simpler to implement
than the other texture-synthesis techniques that have been published
at SIGGRAPH in recent years.
Unfortunately this pixel-centricity of the algorithm means that it
would not be good for high-quality font resizing. Another main
reason that this algorithm would not work for fonts is that the
energy function basically runs the image through an edge detector and
then finds low-energy paths through the resulting image. This proves
to be a good heuristic for photographs, but fonts are pretty much
"all edge" to an edge detector (I'm thinking of normal-sized document
fonts, not large fonts used in eg: an advertisement), so the
heuristic would be unlikely to make good choices about which pixels
to remove.
The basic problem with high-quality pixel-based font resizing in
general (not just this technique) is that font design is a very
difficult skill; experts make a living doing nothing but designing
fonts. These people are artists with a fine sense of balance,
symmetry, etc. who have developed a deep intuition about how slight
changes in a font will be perceived (typically unconsciously) by the
viewer. One pixel can easily make the difference between an
excellent character and complete rubbish. In order to approach this
level of performance, the computer program would need to test
potential font modifications against a model of human visual
perception. Even harder, the model should be as sophisticated (read
"refined", not "complicated") as an expert font designer, not the
average person.
If available, a much better choice is to use an outline-based font
(eg: TrueType) and render it at the desired size.
Josh
On Aug 30, 2007, at 8:15 AM, subbukk wrote:
> On Thursday 30 August 2007 10:55 am, bradallenfuller at yahoo.com wrote:
>> Have you seen this video about image resizing
>> http://www.youtube.com/watch?v=qadw0BRKeMk
> Brad,
>
> This is a very interesting approach to image resizing. The video
> shows only
> photographic images. Has it been applied to font resizing?
>
> Regards .. Subbu
>
More information about the Squeak-dev
mailing list
|