Sunday experiments

Software G'MIC
1

Accidentally, I found a way to generate nice textures :

  • I start with a set of random points, generated with the noise_poissondisk command (I love this one, thanks again @garagecoder !).
600,600 noise_poissondisk. 5

gmic_000003
gmic_000003900Ă—900 11.2 KB

  • Now, for all these isolated points, I look in their neighborhood what other point is the closest one, and I draw a segment between the two.
  100%,100%
  eval.. "i?(
    siz = 2;
    pmin = qmin = dmin = inf;
    while (isinf(dmin),
      y0 = y - siz; y1 = y + siz; x0 = x - siz; x1 = x + siz;
      for (q = y0, q<=y1, ++q,
        for (p = x0, p<=x1, ++p,
          i(p,q) && [p,q]!=[x,y]?(
            d = norm(p - x,q - y);
            d<dmin?(dmin = d; pmin = p; qmin = q);
          );
        );
      );
      siz*=2;
    );
    polygon(#-1,2,x,y,pmin,qmin,1,1);
  )" rm..

gmic_000004
gmic_000004900Ă—900 25.3 KB

  • Now, I apply the close_binary command, that is used in our lineart “smart coloring” filter in the plug-in.
 close_binary. ,

That’s the step that takes most of the computation time. We end up with:

gmic_000005
gmic_000005900Ă—900 37.4 KB

(this command basically tries to connect lines togethers).

  • Now, a bit of anisotropic smoothing and normalization:
  repeat 3 { smooth. 10,0,1,1,2 } n. 0,255 normalize_local. ,

And we get:

gmic_000006
gmic_000006900Ă—900 342 KB

Isn’t that cool ?

And if you combine that with an image, you get something like this:
textured_colorful

which is quite nice.

The whole process is a bit slow because of the close_binary command. But I like the outcome!

That’s it, that was my Sunday experiments :slight_smile:

7 Likes
2

Absolutely splendid!!

1 Like
3

A “spaghetti” variant:

foo :
  srand 0
  600,600 noise_poissondisk. 15

  100%,100%
  eval.. "i?( # For each point, find the closest neighbor.
    siz = 2;
    pmin = qmin = dmin = inf;
    while (isinf(dmin),
      y0 = y - siz; y1 = y + siz; x0 = x - siz; x1 = x + siz;
      for (q = y0, q<=y1, ++q,
        for (p = x0, p<=x1, ++p,
          i(p,q) && [p,q]!=[x,y]?(
            d = norm(p - x,q - y);
            d<dmin?(
              dmin = d; pmin = p; qmin = q;
            );
          );
        );
      );
      siz*=2;
    );
    polygon(#-1,2,x,y,pmin,qmin,1,1);
  )" rm..
  close_binary. 75,1,80,80
  repeat 2 {
    ge. 50% dilate. 2
    repeat 3 { smooth. 10,0,1,1,5 } n. 0,255
  }

spaghetti

2 Likes
4

Now, trying to draw random minimal paths in a graph, using the dijkstra algorithm:

Code:

foo :

  # Create set of vertices.
  600,600 noise_poissondisk. 10 1,1,1,2 eval.. "i?da_push([x,y])" da_freeze. rm.. N:=h => V

  # Create adjacency matrix.
  $N,$N f. ">y>=x?i(y,x):(d = norm(I[#$V,x] - I[#$V,y]); d<20?d:inf)" => A

  # Draw graph.
  600,600,1,3 => canvas
  eval[V] "ellipse(#$canvas,i0,i1,1,1,0,1,0,128,255)"

  # Draw minimal paths between graph nodes.
  repeat inf {
    start,end:=int(u([$N,$N]))%$N
    start:=round($start,round($N/3))
    +dijkstra[A] $start,$end => D
    eval "
      current = $end;
      xyc = I[#$V,current];
      while (current!=$start,
        dist = i(#$D,0,current,0,0);
        isinf(dist)?break();
        parent = i(#$D,0,current,0,1);
        xyp = I[#$V,parent];
        polygon(#$canvas,2,xyc[0],xyc[1],xyp[0],xyp[1],0.03,255);
        current = parent; xyc = xyp;
      );
    "
    rm.
    if !($>%10) w[canvas] filename frame.png,$> o[canvas] ${} fi
  }
4 Likes
5

Fun. More worm or brain-like than pasta noodles, so I wish you a happy belated Halloween.

6

And how was that combination of the spaghetti and the image with withe face done? It looks just cool.

7

I’ve just multiplied the two images together then renormalized in [0,255].

1 Like
8

originalimage.jpg spaghetti.jpg m n 0,255?
The resulting image is nicely daunting :wink: I wonder if the background of the spaghetti image should really be a full black. But I think this is a matter of taste. Happy Sunday, David, and thanks for all the effort.

9

this is my first post here, so please don’t go too hard on me if i’m doing something wrong.

after i saw your amazing picture yesterday, i spent hours trying to achieve the same. i’m a complete noob when it comes to image editing and so i installed gimp and gmic + the gimp gmic plugin…
in order to the first image compiled i put the command snippets together, after i managed to get the command chain right so that my shell doesnt complain about parsing errors, i had my first success and got my first pasta output. now i had to make one image out of 2. i used gimp to multiply them to one pic and “gmic -input multiplied_image.png n 0,255 -output image.png” to normalize it. to be honest the outcome was miles away from yours.
i tried hard but i’m clearly lacking the skills, but i learned something along my way.

can you please explain how you multiplied them together so that the 2 layers look so organic like in your picture, i’m also interested how you achieved the colours?

1 Like
10

Welcome @pixelnoob

Good to hear people are trying this! Without actually having gone through all this myself, I assume it was a multiply of two images in g’mic, on the cli it would be something like:

gmic image1.png image2.png to_rgb mul n 0,255

I added to_rgb because gmic will happily multiply opacity channels as well, so that removes them.

2 Likes
11

thanks for the command garagecoder, that was the command i couldn’t figure out. i just had to make sure that the images to be multiplied had the same size.

after experimenting with it, i come to the conclusion that in order to make the output look organic like in David_Tschumperle’s masterpiece i need to get the ingredients right first. a random combination of inputs doesn’t work well.

i’ll do some more digging!

1 Like