Editing a world map

[Deadshade]

The "Journey Through Texture Space" feature is an excellent way to answer questions of the form "What changes in the image when I change this or that parameter ?" for the fractal functions. You can set a starting and ending value for some parameter and let the system show you what happens as that parameter changes.

Well when I read your short description above, my eyes glazed over But if it does what you say it does, I will increase a bit my time investment to try it. This problem has of course been my biggest difficulty in my Wilbur Learning process.

EDIT : WOW ! I should have tried that MUCH earlier. Much simpler than I first thought and really, really great !! I have already filled up 2 pages of notes of how things change. Just one question - in some cases (f.ex with lacunarity) there seems to be an oscillation like if the image oscilated between a type A and a type B so not a continuous deformation that I expected. Is there a way to show the frames one by one and stop when I would like to look at the map in detail ? Btw I saw nothing when I made the center vary from -10 to 0. Just a random sequence of changing features.

I don't know of good sources for the map elements that you're describing. There may be some here at the guild, but I don't really pay that much attention to such things because that part of mapmaking falls under "artistic" and is well outside of my skill set. You can search for classic mapmakers such as Blaeu or Mercator and look at their maps for elements that are probably in the style of interest. They will still have the problem of being scans, though.

OK. I already did a specific thread about the scale indicators here but apart of 2 examples (apparently self made) I had no luck. Perhaps your idea to search for real maps from old times and extract these elements is the way to go.

[waldronate]

The Journey Through Texture space dialog is pretty much just the fractal function parts of the Compute Height Field dialog with the addition that each parameter has a start and end value (plus a little bit of sequence setup and output things in the lower left).

One of the nice things about trawling old maps is that they often have other frills like border elements and decorative elements that can be used to spice up your maps. One of the easiest ways to apply such things is to adjust the color range so that the new element is black and white and then use a multiply blending mode to just darken the image into the result. This technique works especially well on woodcut style maps, but not so well on the more modern-looking painted maps with lots of saturated colors. I've been trying for years, but have never been able to come up with a good substitute for artistic ability.

[Deadshade]

The Journey Through Texture space dialog is pretty much just the fractal function parts of the Compute Height Field dialog with the addition that each parameter has a start and end value (plus a little bit of sequence setup and output things in the lower left).

.

So I take it that I can't show the frames 1 by 1. Of course the human brain being unable to correctly combine coordinates in an (at least) 5 D parameters space, being able to stop and to take notes before continuing would have helped. I thought that the small square would show a small part of the texture but it only shows black for me.

[Deadshade]

I didn't have to wait a long time to get a kind of graphical glitch I have been talking about. I have been running the Journey for plenty of settings (hundreds) .
This picture was obtained for Hetero terrain as last of the series generated by the Journey (it corresponds to the default setting 1,2,10,1)


As it looked strange, I obtained immediately after that the following one by running Calculate Height with same settings and Hetero terrain :

[waldronate]

Some values of the fractal functions are only useful in certain ranges. Lacunarity is especially sensitive to this. Values much above or below 2 tend to be less than useful. A value of 1.0, for example, just repeats the basis function multiple times. A value of 10 will alias horribly (get high frequencies beyond the function sampling). As lacunarity hits harmonics, it will tend to resemble the lower-frequency harmonics.

Another thing to watch out for is a too-large change of values. If, for example, you move the center of computation from -10 to 0 over 10 frames while viewing an area from -1 to +1, then the image will move so far that your eye can't track it. Using 100 frames or even 1000 frames in this case will make things easier to visually track. You can always hit Cancel on the UI and I think that the system will have the current parameters in the Compute Height Field dialog.

The texture journey UI has a slider that should calculate things on that little black square, but it seems that I broke that somewhere along the way. You can set the system to save the images to disk and it will write the image file sequence as it goes. I was originally interested in producing animations back when machines were very slow, so I would tend to set things to process and leave it alone.

I'm not sure when I last used the Hetero Terrain fractal type. It looks suspiciously like the sphere radii are much larger than they should be (the behavior near the poles suggest that the radii are 30 or more) or that the lacunarity has a strange value. I've never seen this sort of problem before.

[Deadshade]

Still exploring the parameter space. Actually there is little difference between some types regardless of settings.
Especially Multifractal and hybrid multifractal - both are hardly sensible to H values (with the exception of very low ones) and produce very similar structures and not very sexy ones at that. What was your idea to include them in the choices ?

The impact of radius size strongly decreases with the value. Going from 1 to 10 (with my test map size) is about all there is. After 10 the changes are negligible.

I am also still looking about a general way to restore continuities in altitude. As you said that deterracing doesn't really work well if the world is not heavily terraced, it is not the solution.
Is there a way that a selection is filled with a slope which connects on the boundary with the outside height field ? Like mound but just a slope (or a half mound).

[waldronate]

I implemented the basic set of fractal functions in the old Musgrave paper because that was what I had available. I've never used much other than fBm and Ridged Multifractal because they don't offer too much. It's a common problem with software developers: expose what you have other than what's useful.

Remember that the spherical evaluation is effectively inflating a balloon through a cloud and where the balloon intersects the cloud is the basic feature size of the pattern. Probably a bad analogy, but the impact of radius should be continuously-decreased feature size with increasing radius to the point where the features approach single-pixel size and the drawn image looks like random noise. As the radius continues to increase, aliasing of the altitude data for a given image size will make larger features appear sporadically.

I am also still looking about a general way to restore continuities in altitude. As you said that deterracing doesn't really work well if the world is not heavily terraced, it is not the solution.
Is there a way that a selection is filled with a slope which connects on the boundary with the outside height field ? Like mound but just a slope (or a half mound).

I do not understand this question. What problem are you trying to solve (pictures would be helpful).

[Deadshade]

I do not understand this question. What problem are you trying to solve (pictures would be helpful).

I loaded the selection of the original coast. It shows the différences to the coast obtained by the feathering process. There is land in sea and sea in land on wrong places.


Then I eliminate those "wrong" pièces by min and max. I wanted to isolate them first by add and substract from selection but these commands don't seem to work. I obtain the restored coast.


And here is the problem :



There are discontinuities at the interfaces where the min and max applied. So my question was whether it is possible to do something so that the selection in the last image becomes continuous in the vertical dimension (eliminating the "cliffs").
Another problem is the excessive pixelisation of the coastline. Is there a way to do some sort of antialiasing ?

[waldronate]

Some of the observed pixelisation of the coastline is likely due to viewing the surface at 200% zoom as shown in the title bar. View>>Full Size may reduce the effect somewhat.

There's a really ugly trick that can be used to get a selection with a roughly trapezoidal shape and perfect edge. Basically, load a selection, make a mound, clip the mound to flatten the top, smooth the edges, and then apply the noise.

[Deadshade]

Yes this is a great idea that does the job.
Actually in the beginning I tried to use the mound only to create mountains but it doesn't give a good result. Now I realize that it can do many bluffing tricks in editing because it has an interesting property that it creates continuous slopes.

Also I think that with your "cloud" analogy I now at last understood what the center and the radius setting really does. I'll reformulate so that you can tell me if it is right or not.

Let f(x,y,z) be a scalar fractal field defined over the whole space. As a fractal is selfsimilar, the field reproduces similar shapes of a typical size T everywhere.
Let G be an arbitrary sphere of center C and radius R.
The surface S of the sphere is a set of points P(x,y,z) that verify g(x,y,z) = 0. We could choose any arbitrary surface what would just change g but we stay with the sphere.
For every P(x,y,z) belonging to S I sample the values f(x,y,z) which correspond to the intersection of G and the field. These values will define the height field h(x,y,z) on the surface of the sphere G.
Now I translate and scale the sphere G so that it coincides with the planetary sphere and I obtain what I wanted - the height value for every point of the planet's surface.

If what I infer above is right then changing the center C doesn't do much. It just defines another intersection of the sphere and the fractal field so that the h(x,y,z) is distributed in a different way but the structure is inchanged (the fractal field is self similar). Changing the center is then equivalent to just redefine the fractal field by taking another seed and one can forget about it.
However changing R, changes much. The ratio of the typical size of the fractal field to the surface of the sphere is T/R². T is constant for a given field. So when I increase R, this ratio decreases and that means that the typical fractal structure appears smaller compared to the surface of the sphere. But as the derivative of the function 1/R² tends to zero very fast, the relative change of the ratio tends to 0 too. This would be what I observed in a post above that for large R the images look more and more identical when R increases farther.
In other words f.ex when R is 10, to observe the same effect as when one goes from 2 to 3, one would have to increase R from 10 to 47 etc.

(Damn me ! I again succumbed to the temptation to try to understand how things work instead of just staying with the question what qualitative effect a given command has)

EDIT : Btw does the substract from and add to selection command work ? It didn't for me.