Once again I have delved into simulating video feedback.

Here is a 4K resolution 60 fps movie with some samples of what the new simulations can do.

This third attempt is fairly close to second version but with a few changes I will explain here.

The main change is being able to order the effects. This was the idea that got me programming version 3. A shuffle button is also provided that randomly orders the effects. Allowing the effect order to be customised gives a lot of new results compared to the first 2 video feedback simulation modes.

Here are some explanations for the various effects.

HSL Flow

Takes a pixel’s RGB values, converts them into HSL values and then uses the HSL values in formulas to select a new pixel color. For example if the pixel is red RGB(255,0,0), then this converts to HSL(0,1,0.5) assuming all HSL values range from 0 to 1. The length formula above is H*360 and the length formula is s*5. So in this case the new pixel value read would be 5 pixels away at the angle 0 degrees. Changing these formulas allows the image to “flow” depending on the colors.

Sharpen

Sharpens the image by blurring the image twice using different algorithms (in my case I use a QuickBlur (Box Blur) and a weighted convolution kernel). The second blur value is subtracted from the first and then using the following formula the target pixel value is found. newr=trunc(r1+amount*(r1-r2))

Blur

Uses a standard gaussian blur.

Blend

Combines the last “frame” and the current frame. Various blend options change how the layers are combined.

Contrast

Standard image contrast setting. Can also be set for negative contrasts. Uses the following formula for each of the RGB values r=r+trunc((r-128)*amount/100)

Brightness

Standard brightness. Increases or decreases the pixel color RGB values.

Noise

Adds a random value to each pixel. Adding a bit of noise can help stop a simulation dying out to a single color.

Rotate

Guess what this does?

Histogram

Uses a histogram of the image to auto-brightness. Can help the image from getting too dark or too light.

Stretch

Zooms the image. Various options determine the algorithm used to zoom the image.

Image Processing

Allows the various image processing functions in Visions of Chaos to be injected into the mix for even more variety.

As always, you can experiment with the new VF3 mode in the latest version of Visions of Chaos.

I would be interested in seeing any unique results you come up with.

For the next version 4 simulation I would like to chain various GLSL shaders together that make the blends, blurs etc. That will allow the user to fully customise the simulation and insert new effects that I did not even consider. Also GLSL for speed. Rendering the above movie frames took days at 4K resolution.

One of the features I have wanted to implement since the earliest versions of Visions of Chaos has been a formula editor and compiler so users can experiment with their own fractal formulas. This has also been requested many times from various users over the years. Now it is finally possible.

Rather than write my own formula parser and compiler I am using the OpenGL Shading Language. GLSL gvies faster results than any compiler I could code by hand and has an existing well documented syntax. The editor is a full color syntax editor with error highlighting.

As long as your graphics card GPU and drivers support OpenGL v4 and above you are good to go. Make sure that you have your video card drivers up to date. Old drivers can lead to poor performance and/or lack of support for new OpenGL features. In the past I have had outdated drivers produce corrupted display outputs and even hang the PC running GLSL shader code. Always make sure your video drivers are up to date.

For an HD image (1920×1080 resolution) a Mandelbrot fractal zoomed in at 19,000,000,000 x magnification took 36 seconds on CPU (Intel i7-4770) vs 6 seconds on GPU (GTX 750 Ti). A 4K Mandelbrot image (3840×2160 resolution) at 12,000,000,000 x magnification took 2 minutes and 3 seconds on CPU (Intel i7-6800) and 2.5 seconds on GPU (GTX 1080). The ability to quickly render 8K res images for dual monitor 4K displays in seconds is really nice. Zooming into a Mandelbrot with minimal delays for image redraws really increases the fluidity of exploration.

So far I have included the following fractal formulas with the new Visions of Chaos. All of these sample images can be clicked to open full 4K resolution images.

Buffalo Fractal Power 2

Buffalo Fractal Power 3

Buffalo Fractal Power 4

Buffalo Fractal Power 5

Burning Ship Fractal Power 2

Burning Ship Fractal Power 3

Burning Ship Fractal Power 4

Burning Ship Fractal Power 5

Celtic Buffalo Fractal Power 4 Mandelbar

Celtic Buffalo Fractal Power 5 Mandelbar

Celtic Burning Ship Fractal Power 4

Celtic Mandelbar Fractal Power 2

Celtic Mandelbrot Fractal Power 2

Celtic Heart Mandelbrot Fractal Power 2

Heart Mandelbrot Fractal Power 2

Lyapunov Fractals

Magnetic Pendulum

Mandelbar (Tricorn) Fractal Power 2

Mandelbar Fractal Power 3

Mandelbar Fractal Power 3 Diagonal

Mandelbar Fractal Power 4

Mandelbar Fractal Power 5 Horizontal

Mandelbar Fractal Power 5 Vertical

Mandelbrot Fractal Power 2

Mandelbrot Fractal Power 3

Mandelbrot Fractal Power 4

Mandelbrot Fractal Power 5

Partial Buffalo Fractal Power 3 Imaginary

Partial Buffalo Fractal Power 3 Real Celtic

Partial Buffalo Fractal Power 4 Imaginary

Partial Burning Ship Fractal Power 3 Imageinary

Partial Burning Ship Fractal Power 3 Real

Partial Burning Ship Fractal Power 4 Imageinary

Partial Burning Ship Fractal Power 4 Real

Partial Burning Ship Fractal Power 5

Partial Burning Ship Fractal Power 5 Mandelbar

Partial Celtic Buffalo Fractal Power 4 Real

Partial Celtic Buffalo Fractal Power 5

Partial Celtic Burning Ship Fractal Power 4 Imaginary

Partial Celtic Burning Ship Fractal Power 4 Real

Partial Celtic Burning Ship Fractal Power 4 Real Mandelbar

Perpendicular Buffalo Fractal Power 2

Perpendicular Burning Ship Fractal Power 2

Perpendicular Celtic Mandelbar Fractal Power 2

Perpendicular Mandelbrot Fractal Power 2

Quasi Burning Ship Fractal Power 3

Quasi Burning Ship Fractal Power 5 Hybrid

Quasi Celtic Heart Mandelbrot Fractal Power 4 Real

Quasi Celtic Heart Mandelbrot Fractal Power 4 False

Quasi Celtic Perpendicular Mandelbrot Fractal Power 4 False

Quasi Celtic Perpendicular Mandelbrot Fractal Power 4 Real

Quasi Heart Mandelbrot Fractal Power 3

Quasi Heart Mandelbrot Fractal Power 4 Real

Quasi Heart Mandelbrot Fractal Power 4 False

Quasi Heart Mandelbrot Fractal Power 5

Quasi Perpendicular Burning Ship Fractal Power 3

Quasi Perpendicular Burning Ship Fractal Power 4 Real

Quasi Perpendicular Celtic Heart Mandelbrot Fractal Power 4 Imaginary

Quasi Perpendicular Heart Mandelbrot Fractal Power 4 Imaginary

Quasi Perpendicular Mandelbrot Fractal Power 4 False

Quasi Perpendicular Mandelbrot Fractal Power 5

A lot of the above formulas came from these summaries stardust4ever created.

All of these new custom fractals are fully zoomable to the limit of double precision floating point variables (around 1,000,000,000,000 magnification in Visions of Chaos). The formula compiler is fully supported for movie scripts so you can make your own movies zooming into these new fractals.

This next sample movie is 4K resolution at 60 fps. Each pixel was supersampled as the average of 4 subpixels. This took only a few hours to render. The resulting Xvid AVI was 30 GB before uploading to YouTube.

So, if you have been waiting to program your own fractal formulas in Visions of Chaos you now can. I look forward to seeing the custom fractal formulas Visions of Chaos users create. If you do not understand the OpenGL shading language you can still have fun with all the default sample fractal formulas above without doing any coding.

DLA is a simple process to simulate in the computer and was one of my first simulations that I ever coded. For the simplest example we can consider a 2D scenario that follows the following rules;

1. Take a grid of pixels and clear all pixels to white.
2. Make the center pixel black.
3. Pick a random edge point on any of the 4 edges.
4. Create a particle at that edge point.
5. Move the particle randomly 1 pixel in any of the 8 directions (N,S,E,W,NE,NW,SE,SW). This is the “diffusion”.
6. If the particle goes off the screen kill it and go to step 4 to create a new particle.
7. If the particle moves next to the center pixel it gets stuck to it and stays there. This is the “aggregation”. A new particle is then started.
Repeat this process as long as necessary (usually until the DLA structure grows enough to touch the edge of the screen).

When I asked some of my “non-nerd” aquaintances what pattern would result from this they all said a random blob like pattern. In reality though the pattern produced is far from a blob and more of a coral like branched dendritic structure. When you think about it for a minute it does make perfect sense. Because the moving particles are all coming from outside the fixed structure they have less of a chance of randomly walking between branches and getting to the inner parts of the DLA structure.

All of the following sample images can be clicked to see them at full 4K resolution.

Here is an image created by using the steps listed above. Starting from a single centered black pixel, random moving pixels are launched from the edge of the screen and randomly move around until they either stick to the existing structure or move off the screen.

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: when they move next to a single stuck particle

The change in this next example is that neighbouring pixels are only checked in the N, S, E and W directions. This leads to horizonal and vertical shaped branching.

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W
Particles stick condition: when they move next to a single stuck particle

There are many alternate ways to tweak the rules when the DLA simulation is running.

Paul Bourke’s DLA page introduces a “stickiness” probability. For this you select a probability of a particle sticking to the existing DLA structure when it touches it.

Decreasing probability of a particle sticking leads to more clumpy or hairy structures.

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: 0.5 chance of sticking when they move next to a single stuck particle

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: 0.1 chance of sticking when they move next to a single stuck particle

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: 0.01 chance of sticking when they move next to a single stuck particle

Another method is to have a set count of hits before a particle sticks. This gives similar results to using a probability.

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has hit stuck particles 15 times

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has hit stuck particles 30 times

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has hit stuck particles 100 times

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has hit stuck particles 1000 times

Another change to experiment with is having a threshold neighbour count random particles have to have before sticking. For example in this next image all 8 neighbours were checked and only 1 hit was required to stick, but the moving particles needed to have at least 2 fixed neighbour particles before they stuck to the existing structure.

The first 50 hits only need 1 neighbour. This build a small starting structure which all the following hits stick to with 2 or more neighbours.

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has 2 or more stuck neighbours

That gives a more clumpy effect without the extra “hairyness” of just adding more hit counts before sticking.

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has 3 or more stuck neighbours

This seems a happy balance to get thick branches yet not too fuzzy ones.

Start configuration: Single centered black pixel
Random pixels launched from: rectangle surrounding existing structure
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has 3 or more stuck neighbours and has hit 3 times

The next two examples launch particles from the middle of the screen with the screen edges set as stationary particles.

Start configuration: Edges of screen set to black pixels
Random pixels launched from: middle of the screen
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has 1 or more stuck neighbours

Start configuration: Edges of screen set to black pixels
Random pixels launched from: middle of the screen
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has 3 or more stuck neighbours

These next three use a circle as the target area. Particles are launched from the center of the screen.

Start configuration: Circle of black pixels
Random pixels launched from: middle of the screen
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has 1 or more stuck neighbours

Start configuration: Circle of black pixels
Random pixels launched from: middle of the screen
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has 2 or more stuck neighbours

Start configuration: Circle of black pixels
Random pixels launched from: middle of the screen
Neighbours checked for an existing pixel: N, S, E, W, NE, NW, SE, SW
Particles stick condition: once a moving particle has 3 or more stuck neighbours

This next image is a vertical 2D DLA. Particles drop from the top of the screen and attach to existing seed particles along the bottom of the screen. Color of sticking particles uses the color of the particle they stick to.

Start configuration: Bottom of screen is fixed particles
Random pixels launched from: top of the screen
Neighbours checked for an existing pixel: S, SE, SW
Particles stick condition: once a moving particle has 1 or more stuck neighbours

Dendron Diffusion-limited Aggregation

Another DLA method is Golan Levin’s Dendron DLA. He kindly provided java source code so I could experiment with Dendron myself. Here are a few sample images.

Hopefully all of the above examples shows the wide variety of 2D DLA possible and the various methods that can produce them. This post was mainly going to be about 3D DLA, but I got carried away trying all the 2D possibilities above.

3D Diffusion-limited Aggregation

DLA can be extended into 3 dimensions. Launch the random particles from a sphere rather than a surroundiong circle. Check for a hit against the 27 possible neighbours for each 3D grid cell.

For these movies I used default software based OpenGL spheres. Nothing fancy. Slow, but still managable with some overnight renders.

In this first sample movie particles stuck once they had 6 neighbours. In the end their are approximately 4 million total spheres making up the DLA structure.

This second example needed 25 hits before a particle stuck and a particle needed 5 neighbours to be considered stuck. This results in a much more dense coral like structure.

A Quick Note About Ambient Occlusion

Ambient occlusion is when areas inside nooks and crannies are shaded darker because light would have more trouble reaching them. Normally to get accurate ambient occlusion you raytrace a hemisphere of rays from the surface point seeing how many objects the rays intersect. The more hits, the darker the shading. This is a slow process. For the previous 3D DLA movies I cheated and used a method of counting how many neighbour particles the current sphere has. The more neighbours the darker the sphere was shaded. This gives a good enough fake AO.

Programming Tip

A speedup some people do not realise is when launching random particles. Use a circle or rectangle just slightly larger than the current DLA structure. No need to start random points far from the structure.

Other Diffusion-limited Aggregation Links

For many years now the ultimate inspiration in DLA for me has been Andy Lomas.

Another change Andy uses is to launch the random walker particles in patterns that influence the structure shape as in this next image.

Future Changes

1. More particles. The above 2 movies have between 4 and 5 million particle spheres at the end of the movies. Even more particles would give more natural looking structures like the Andy Lomas examples.

2. Better rendering. Using software (non-GPU) OpenGL works but it is slow. Getting a real ray traced result supporting real ambient occlusion and global illumination would be more impressive.

3. Controlled launching of particles to create different structure shapes other than a growing sphere.

Mobile phones and tablets have had nice high DPI displays for a long time now. PC screens have started to catch up and 4K monitors are starting to come down in price and will become the standard in the future.

The good thing about higher DPI screens is that there are many more pixels on the same sized screen so text and images are shown much more clearly.

The bad thing is most applications (especially older programs) are not aware of High DPI scaling settings. If an application does not support scaling Windows just “zooms” the application window to a larger size. This leads to blurry text that most people complain about. So you either set the scaling factor to 100% which leads to tiny text or you live with blurry text.

Over the past few weeks I have been working on making Visions of Chaos high DPI aware. The end result is that it will now run happily on all scale settings and higher 4K monitors without the text being too tiny to read or blown up and blurry. The menus and all text within Visions of Chaos will scale according to your Windows settings and remain smooth and crisp to read.

High DPI in Visions of Chaos

Prior to Visions of Chaos being DPI aware Windows would just scale the output window like in a photo editor app. The following shows how the Visions of Chaos window looks at 150% scaling under Windows 10 on a 4K display.

This leads to the blurred text and the generated image (set to 480×360 pixels in size) is shown onscreen also stretched and blurred to 725 pixels wide.

Here is the same window using the latest version of Visions of Chaos that is DPI aware. The text fonts are scaled to a larger size so they remain crisp and smooth. The fractal image that has been set to 480×360 pixels is exactly 480×360 monitor pixels.

Windows is still playing catch up with higher resolution DPI support, but it will hopefully get better as 4K monitors become the standard. At least you can be confident that Visions of Chaos will continue to be usable and readable at high DPI settings.

Now I can get back to adding new features to Visions of Chaos.

It took a lot longer than I wanted, but the new Visions Of Chaos for Android v1.0 is now available for download. It is 100% free and ad free.

The main new feature is support for running OpenGL ES shaders on your Android device. There are over 1000 sample shaders included with the app.

So, if you own one of the 85% market share Android devices download the app from the Visions Of Chaos for Android web page or directly from Google Play now.

Back in 2007 I was trying to figure out the algorithms behind Jonathan McCabe’s cellular automata pictures. I was never able to determine the algorithm to get the same results and coloring as Jonathan, but this is an explanation of my own Coupled Cellular Automata that arose from trying.

Algorithms

The principal of this method is to have 3 CAs with different settings running mostly independently of each other, but have their values interact with each other to create a new interesting result.

To start with you need 3 arrays to keep track of the 3 separate CAs. I used a 3D array to store them, ie c[0,x,y] c[1,x,y] and c[2,x,y].

Each cell is updated using a kernel which is very similar to an image processing kernel. For example, this kernel

is like a weighted blur image processing kernel.

Initialise a value to 0. The current cell value is multiplied by 4 and added to the total. The north, south, east and west neighbour cell values are mutlipled by 2 and added to the total. The corner cell values are mutliplied by 1 and added to the total.

Depending on which of the 3 CA layers is being processed, the value is updated by

case level of
0:value:=value div k1total-(c[2,x,y] mod k1mod+1);
1:value:=value div k2total-(c[0,x,y] mod k2mod+1);
2:value:=value div k3total-(c[1,x,y] mod k3mod+1);
end;

k1total, k2total and k3total are the total of all the kernel values. For the above kernel the total value is 16 (4+2+2+2+2+1+1+1+1).

k1mod, k2mod and k3mod are the 3 “Mod value” settings that can be customised.

The value is also modified by using the cell value in another layer. This gets the CAs feeding through each other and causes the multiple layered look of the CA.

Once you have the new value for the cell, it is then kept within the range of 0 to 255 by the Wrap Method. The 3 wrap methods (Mod, Inverse and Clamp) make the following changes to the value

case wrapmethod of
0:value:=value mod 255;
1:if value > 255 then value:=-255 else if value < -255 then value:=255;
2:if value > 255 then value:=255 else if value < -255 then value:=-255;
end;

At this stage you now have a new value for the cell. Put this cell into a temp array and then swap the temp array into the cell array once all cells are processed (like any other CA you want to update all the cells simultaneously).

Repeat the above steps as long as you need to.

Display Methods

You now have the 3 CA layers updated with their new values. I use 3 different methods to display the CA.

1. Color Palette. Index a 256 color palette with abs(c[0,x,y]+c[1,x,y]+c[2,x,y] div 3) mod 255;

2. RGB. Convert the layer values to a color by using RGB(abs(c[0,x,y]),abs(c[1,x,y]),abs(c[2,x,y]));

3. YUV. Pass the layer values as YUV into a YUV to RGB converter routine YUVtoRGB(abs(c[0,x,y]),abs(c[1,x,y])-128,abs(c[2,x,y]-128),r1,g1,b1);

Results

Here are some more example kernels and the resulting images. The settings dialog screen captures also show off the new skinning theme support that Visions Of Chaos now includes.

Of course like any CA these need to be seen dynamically changing and updating. To see the above samples and others in motion download Visions Of Chaos.

If you are a coder who experiments with the above algorithms and has any other ideas or enhancements let me know.