Image to masked PF WIP

[bigben]

[Ahh crap... Just saw the image was still loaded underneath]

Here's a WIP of my experiments with using colour similarity masks within TG to convert a texture image to a series of masked PFs.  This is 4 colours... each colour consisting of a masked surface of the reference colour and a single child layer of a lighter colour with low coverage and fractal breakup. Extra variation to the colours comes from the overlapping of masks.  From some earlier experiments I think I'll need somewhere between 12 - 16 colours.

Still some a lot finessing to do with very light and dark colours, and learning how to incorporate the lightness comparison more effectively but it's showing some promise.

[bigben]

I was obviously suffering from TOS when I set this up the first time and had overcomplicated the mask network, adding two unnecessary calculations which was making it difficult to correlate the output values to the desired mask... and it was all downhill from there. So here is a proper WIP image. First is 4 masked surfaces (dark green, brown, orange-brown and "sand"), second image is the image map only.

Pays to also have a rendered image map view for reference looking at this I thought the green was spreading too much and cut it back

I'm selecting the reference colours from the raw image in Photoshop, is in TG I can remove the colour and saturation adjust nodes (and set the colour of the image to Linear so that the reference colours match the image)  Much happier now, but other things to get on with this weekend.

[bobbystahr]

Cetainly happy to see someone with full colour sense working this out. As a functional colour blind artis t(I use an  R,G,B table I've had since the '90's for selecting if a picker option isn't present) I'd not stand a chance. I would miss the subtly of the tones. Great promise Ben.

[Mahnmut]

Hi Ben,
I am not sure I understood this right:
Does TG have a function to make a mask from the color of an image, or do you do that in a graphics program?
I tried the latter some time ago, using my huge green mars map. First I converted it to 16 colors, than selected all pixels of one color, made them white on black and blurred them slightly. That gave me 16 huge greyscale maps of the colors´ distribution, but I never  really succeeded in using these as distribution masks.
So is there a smarter way?
There is some hope I will be able to spend some more time on TG again, since I am in a less life-consuming job since march 1st .
Cheers,
J

[Oshyan]

Ben is figuring out how to do color-range masking *in TG*. So when he publishes the resulting files/networks, I'm sure you can adapt it easily to Mars, which would be very helpful for getting proper detail nearer ground level in the animation, eh?

- Oshyan

[bigben]

Yes, that's exactly the way I used to do it.  Load a Landsat image into Photoshop and create multiple greyscale masks for each of the features (forests, fields etc...), taking up most of your RAM with large images. 

This approach does the colour range selection using TG functions, so you only need to load the source reference image.  You choose a reference colour (I get the RGBs via the colour picker in Photoshop) and it spits out a colour range selection mask and a lightness difference map (might have to change that to a subtraction) for the image. These are used to mask TG surface shaders with colours of your choosing (no need to correct the colour in false colour Landsat images before loading for example)

It looks something like this (prior to cleaning up ) so those renders are entirely TG surfaces.  It takes a bit of practice getting used to working with them but I'm pretty excited by how it's coming together.

[bobbystahr]

Yes, that's exactly the way I used to do it.  Load a Landsat image into Photoshop and create multiple greyscale masks for each of the features (forests, fields etc...), taking up most of your RAM with large images. 

This approach does the colour range selection using TG functions, so you only need to load the source reference image.  You choose a reference colour (I get the RGBs via the colour picker in Photoshop) and it spits out a colour range selection mask and a lightness difference map (might have to change that to a subtraction) for the image. These are used to mask TG surface shaders with colours of your choosing (no need to correct the colour in false colour Landsat images before loading for example)

It looks something like this (prior to cleaning up ) so those renders are entirely TG surfaces.  It takes a bit of practice getting used to working with them but I'm pretty excited by how it's coming together.

Freakin brilliant...

[Dune]

Obviously a clever mind at work! Great thinking, Ben!

[TheBadger]

That network is intimidating

[bigben]

That network is intimidating

It's just laid out a bit awkwardly to save scrolling the screen too much, but at its core it's just the same old colour shader with a distribution mask. 

Having buried most of the math, I went through to double check the lightness mask: Colour to vector -> Length to scalar > divide by square root of 3 .... and then .... oh wait, what this new fangled "Colour to greyscale colour" function     Gotta read me some more release notes.

[bigben]

More of a proof of concept than something pretty. Adding in some breakup within each colour still leaves you with a relatively soft image.  In this test, a PF is masked by the complement of the original mask and then subtracted from it. This image is 2 masked colours on a black background, with adjacent segments of the original image map.

[Oshyan]

Interesting and a bit extreme. I think I'd want to see it with less Warp, and perhaps having more cohesion between even differently colored textures (e.g. same Seed but different Settings in the Power Fractal) would make it look more natural. Promising though.

- Oshyan

[bigben]

The same seed for different colours looked odd, although alternating between a PF and its complement may work, as well as reducing the potential for gaps. I agree about the PF settings, but I find at this stage that they help me see when other settings actually have an impact.  In a 4 colour setup I was tweaking one colour for about 15 minutes, only to find that when I made the other 3 invisible that the surface was barely visible... 

Next cab off the rank will be going back to high contrast masks and trying to come up with a basic palette of shaders that will give a good approximation of the image texture and then return to this to break up the boundaries. 

The other thing I found with this approach is that because I'm not using a "black" shader, many of the lakes in the Blue Marble image are a different colour, so the image offset is less of a problem. As the rivers/lakes distribution is controlled by an image mask I could make up a water-side mask which could be used to disguise this further.

[bigben]

It's still a little quirky to use in that you need to preview the masks while tweaking the settings, but I think I'm ready to start adding in the rest of the colours.  For each colour I have 2 child layers for colour variation; one darker, one lighter both masked by their relative lightness.
Picked up some tips for the PFs from NVSeal http://www.planetside.co.uk/forums/index.php/topic,2747.msg27929.html

Single colour on a black background and adjacent portions of image map.

[Mahnmut]

So the bright tiles in your image show the image map and the dark tiles the masked orange layer on black background?
Looks good so far.