Challenge: Stippled outline

Didn't Corvus do something like this in his junkyard a while back??

Just looked through the whole thing (impressive stuff in there!) but only the recent pages look related and then again its not the same. I'm getting close with the bomber - the main problem using the bomber is that it might drop an extra particle where it shouldn't (or skip one)

Edit: this image shows the problem

Quote
Here's a (slightly crazy) idea: Using the new derivative command, you can calculate the slope, which can give you the correct rotation. (like a normal map) But it wouldn't tell you how to place the dashes evenly.

True, I played with this idea even before we had the new math components. Theoretically one should be able to find the perpendicular slope and use that in further processing.

Here is an early test with the new derivative component. First the shape is expanded via distance transform, then I find the normals and map an angular gradient. This gives the orthogonal vector to the expansion. These two are then hooked up to the new Lookup along with my stippling shape.

The main problem is the changing frequencies and "warping" due to pre smoothening of the distance transform (necessary to reduce aliasing)

Still some problems to solve, but at least you can see the shape of the stipple particle now

Huh... tried to duplicate your ortho vector thing. Wound up with this instead:
Not really useful, but interesting looking.

Quote
CorvusCroax wrote: 1) You lost me at 'orthagonal vector'. Want to post one of your working filters and I (or others) could take a look and help out? 2) looking at your second image: you've got an x and a y version of this you're feeding into lookup right? I wonder if you couldn't use tone curves just on that to creat dashed lines. If all you want is dashes... i think you've just about got it.

I didn't loose you completely Yes, that second last image I posted is one out of two needed to map the stipple particle. The other is the distance transform (it could be a blurred version of the input too, but it is not very precise).

To map an image somewhere you could simply move it there and rotate and scale. Thats what I am doing with several "copies", I just only consider the coordinate space, not the actual image (i.e. what goes to the new Lookup's x and y inputs).

So with a path we have two mapping directions we need to construct, one that "point" outwards, perpendicular to the outline path (I use the distance transform here), and one direction in parallel with the outline path (the tough one I'm trying to perfect here). I might not be very clear in the use of terms here, sorry When I speak of the orthogonal direction it is in relation to the distance transform (or outgoing direction), not the actual path direction. I'll refer to it as the parallel direction from now on.

Once you have these two x,y mapping inputs you can map any type of stipple particle, not just simple dot stippling. The reason I asked about simple stippling initially was to simplify the explanation of the problem.

Quote
CorvusCroax wrote: 3) high pass! High pass can help you filter out your high frequency noise areas from your low freq areas (that is, the tight spots vs the stretched spots.)

Been there already, but perhaps I should revisit that approach. The frequency curve op can change the frequency of repeating linear curves (via a repeat op first). I'm trying to figure out a way to calculate the given frequency at given a point in the parallel direction. The Derivative component is useful here, but the result I get is currently too damaged..

I'll get back with some more info and snippets later

Here is an attempt to figure out the frequency of a given area and then relative to that try to compensate.

The problem here is that the original span shifts do not align with the compensation frequency adjustments (causing new small spans or skipping).



Frequency Adjust.ffxml

And the filter:

Ants 5.ffxml

Expanded ants. Looks a lot more uniform.

Ooh definitely getting there.

Sphinx: this is some impressive work! Thanks for posting the snippets.

The core derivative -> lookup part is genius.
I noticed that the hues seem to be slightly compressed and uneven in some areas, and tried to fix it, but ran hard into my own ignorance. (eg: Plugged in a (regular) gradient and it just doesn't work at all.) Could you explain that bit? Is it important that the derivative angles are slightly off? Why not use a normals construction similar to Mikes 'RealNormal mapper'?

Quote
Sphinx. wrote: Expanded ants. Looks a lot more uniform

Looking good!

The reason why the lookup works is because the derivative values are like .00001233 and 20. (yaay unclamped workflow!) The free gradient up there goes off into infinity, and the Lookup is pulling values from well off the 'frame' of the gradient.

Fundamentally, though, the Ants method is probably better, since this method, although it is very interesting, is always going to divide curves into fixed radial divisions.

The cool thing, though, is that by tweaking the curve profile of the free gradient, you can create different dot patterns. (dash-dot-dash,etc)

Here's the file:

Basic Angular Expansion X01.ffxml

I can't believe this stuff didn't "get stoled" yet. No "Stipples & Stitches" filter on the library?!

*** hauls stolen bag-O-tech to secret hideout ***