Abstract
I have just made new algorithm, I called TCNN (Trace Contour Nearest Neighbor).
It is the loop, that contains tree steps:
1. Tracing the shapes with 1-pixel contours and contours of this contours as many times, that the whole image is filled with it.
2. Changing the size of the image with nearest neighbor mode (in this case proportionally 50%) to result flat color in place of complex multi-contour structure.
3. Returning previous size of the image also with nearest neighbor mode.
I haven't examined it in more than 1100 generations yet, but it seems that the process could go to infinity. I also discovered, that for transformation of this particular base image (diagonally asymetric colors) there is some moment in wich it becomes actual symetric. I cropped the composition to this frame.
The TCNN algorithm animation.
modular_layer_representation_n2_l3_8x8
- The base image
modular layer representation_n2_l3_8x8_tcnn200-50 1-1081_scaled0,5nn_part1.gif
The TCNN transformation in pure form. (part 1)
modular layer representation_n2_l3_8x8_tcnn200-50 1-1081_scaled0,5nn_part2.gif
The TCNN transformation in pure form. (part 2)
modular layer representation_n2_l3_8x8_tcnn200-50 1-1081_scaled0,25nn.gif
The TCNN transformation in pure form resized again 50% with nearest neighbor mode.
selfxor_timestrech_modular layer representation_n2_l3_8x8_tcnn200-50 1-1081_scaled0,5nn.gif
The animation shows the differences in color between two sides of the animation (lighted pixels). It is seen, that the sides are getting more and more similar and become exact the same.
move-y1px,x1px_2frame_delay_xor_modular layer representation_n2_l3_8x8_tcnn200-50 1-1081_scaled0,5nn.gif
The differences between actual TCNN animation and its duplication delayed 2 frames in time and 1 pixel up and left. It shows very clearly the most dynamic parts of the process - when a little distortion results fractalic reaction.
