On 06/25/2007 05:23:01 PM, bulia byak wrote:

On 6/24/07, bulia byak <buliabyak@...400...> wrote:

...

On 6/23/07, Tom Lechner <lechner@...1707...> wrote:

...

http://www.iro.umontreal.ca/~ostrom/publications/pdf/SIGGRAPH99_FacialEngrav...

Looks like my enthusiasm was premature. After reading the entire paper, I see that the most difficult part - separating the image into areas and choosing the shape and direction of engraving lines - is still done entirely manually in this approach.....

Can potrace or some such tracer be used to trace out regions that break along a certain kind of differing contrast? In the head sculpture in the paper, the face is a broad area with a gradual change of contrast, which borders things like the nose, hair and background. The borders between these things are much more abrupt changes in contrast. I wonder if there's some way to distinguish regions based on the amount of "tangle" inside (hair versus forehead, for instance)? Group according to certain threshholds of standard deviation of contrast per area?

Whenever I do lots of line work I tend to align lines along contours of darkness, or a little diagonally from it in roundish shapes like people, and perpendicular to contours for blocky areas that should really stand out. Lines on opposite sides of a border between regions might be perpendicular, which emphasizes foreground versus background. This sort of general rule could be used to help define line directions within regions.

My ideal engraving tool would be half automatic and half manual. Regions and general directions for lines could be found automatically, then these guides could then be tinkered with by the artist, who finally decides when to actually turn the guides into engraving lines. Computers can play chess pretty well, but they still don't know much about aesthetics! That Strokes Maker program, which looks kind of neat: http://www.vectools.com/en/StrokesMaker1.aspx seems to let you define a single type of path to mimic. This would at least be comparatively easy to implement: Select a path for a mask, select another path to mimic, and press the "engrave" button.

So anyway, Tom ----------- http://www.tomlechner.com http://www.laidout.org

Bob Jamison wrote:

...

Well, of course we have something like this already. It's called SIOX. Its purpose is to extract regions with a similar color signature.

http://www.siox.org/

If we can apply SIOX to other algorithms, are we inventing something?

...which reminds me of something. A year ago, I promised Hr. Friedland that I would make a GUI that would handle the SIOX algorithm properly. Is there anyone who needs a task to do that might actually be fun?

bob

On 6/27/07, jtaber <jtaber@...480...> wrote:

Bob Jamison wrote:

...

Bob Jamison wrote:

...

...

Well, of course we have something like this already. It's called SIOX. Its purpose is to extract regions with a similar color signature.

http://www.siox.org/

If we can apply SIOX to other algorithms, are we inventing something?

...which reminds me of something. A year ago, I promised Hr. Friedland that I would make a GUI that would handle the SIOX algorithm properly. Is there anyone who needs a task to do that might actually be fun?

I've been playing with converting geographical maps using potrace but have not used siox.

I tried tracing a vidcap frame from Futurama. I wasn't able to get the tracing stuff in inkscape to do a reasonable job at all, and that includes SIOX. It seems to be a hard case for Inkscape because the colors aren't exactly uniform tone over a region due to encoding noise.

There's a lot of applications that can go along

with this - maybe some are more fun than others. Perhaps the biggest problem when using photos is shading is a different color zone but not necessarily a different object (ie a road can have different levels of shading but it still should be identified as the same road). John

The Manga Colorization paper at Siggraph last year had an interesting technique based on I think Gabor wavelet filters to detect regions of similar pattern in an input image. They probably borrowed the core technique from image processing, but that'd be where I'd start looking for how to handle the roadway case.

--bb