Coloring Regions

Some examples of other fill modes...

COLOR RANDOMLY

COLOR BY ORIENTATION

IMAGE COLOR

The reference image in the fill mode examples above is the result of a bricks component piped through a kaleidoscope. While I could replicate a few of the Regions node's fill mode results with a similar setup in FF (I omitted those here), the above six examples would probably require a custom approach for each and/or prove very difficult to do.

I am totally scratching my head over how to do best fit image like stuff in FF. That's just too cool for generating height maps from kaleidoscopic patterns.

Try doing the below example in FF, which is mainly created easily in Genetica due to a Regions node set to Best Fit Image providing the heightmap. The heightmap would be extremely challenging to recreate exactly with its smooth curved folds and hard paper edges, although it might be possible to fake the look with 2D techniques. Also note the accurate shadows processed from the heightmap.

Quote
Crapadilla wrote: Regions are detected by finding the continuous areas

What's really interesting about this is how exactly does it find them -- a pixel-based algo like that of Photoshop's Magic Wand, or something based on the derivative?

Quote
Vladimir Golovin wrote: how exactly does it find them

Looks like a pixel-based algorithm to me, similar to a luma key or wand.

Another possibility would be that all procedural generator nodes pass region IDs down the process stream which tools pick up and modify or act upon. But then again that would probably imply a vector based approach similar to Allegorithmic, and I think Genetica is pixel-based.

Quote
Vladimir Golovin wrote: it uses a sample-based architecture

If that's the case, then I am in high hopes of seeing something like a regions component in Filter Forge's not too distant future. I know you guys can do it...

Quote
CFandM wrote: Hmm, sounds like an advanced filter challange....

Indeed. And after more than a year playing around with FF, I still haven't found a way to do these...

Can we say 'gap in the toolset' together?

Quote
Crapadilla wrote: Can we say 'gap in the toolset' together? Wink Devil Grin

gap in the toolset

Nope I have not found any way as of yet...

The red channel of the lifesaver bitmap put through a Threshold component in FF, Threshold value 90:

Keeping all controls the same, I went through the various fill modes to see how they react on the image...

'Color Radially':

'Color by Orientation':

'Image Color':

'Scaled Image':

I'm guessing the Regions node computes the X/Y dimensions of each region and scales the input image accordingly here...

'Best Fit Image' again, but this time I activated the 'Rotate Image' option.

It appears the Regions node also computes the longest axis of each region and is thus able to align the input image to this axis. Note that the input image is also scaled non-uniformly to fit into each region.

I'd wager the Regions node computes some sort of minimal bounding box for each region, and derives from it the center and the axes of orientation.

Yes, Genetica's regions are a unique feature I haven't seen anywhere else. I considered similar ideas way back, in the pre-beta days. The problem with the regions is that they cannot be implemented procedurally (this would require enormous computation costs), so we'd have to use a bitmap-based component for that. And even with a bitmap-based component, we'd have to deal with fill aliasing issues (i.e. jaggy edges on the region boundaries).

(I had in mind a slightly different idea for regions. Some time ago I found a very nice algorithm for picture segmentation, but unfortunately it was patented. I tried to contact its owners multiple times to discuss licensing, but they never replied.)

Yep, that's correct.

Yes, it's a pity -- that visual segmentation algo was really neat, quite possibly the best of its kind.

Regarding 'attention-worthiness', things are not as gloomy as you imagine -- I had email exchanges and voice conversations with people like Ken Perlin, Kai Krause, John Nack and Sumit Gupta, some of them even before we released the beta -- and they all have been very nice and helpful. So I guess it depends on the situation.

Quote
Crapadilla wrote: Any other industry-biggies you're keeping from us?

No thanks, after the previous post I already look like a namedropper

I don't know for sure -- I haven't seen the algo itself, only the results. Even if it did, the results were great.

UPD: of course it did, at least at some pre-computation stage. I'm just not sure how it stored extracted regions.

One of my most favorite feature request threads is already way beyond its 5th anniversary, so I'm practically forced to bump it.

The question above still stands!

I read that as:

Quote
"FF might quickly become 'feature-complete' in regard to region-based gradients once we release Bitmap Scripts, and the community jumps to the occasion by creating their own scripted implementation(s)."