Example: *P**roof of circle*

(Album cover <https://www.youtube.com/watch?v=dRjJogNRMSY&ab_channel=Mr_Kenzo> 
I used for inspiration - and it's nice music too)

*Let's say you want to convince someone that there is a circle in an image.*

Then you would convey to them something like this:

1. Some preprocessing of the image

Process the image like this:

  pixel = if pixel is 1% (or other value) away from black
    then black else white
  end

2. Creation of the circle pattern

Calculate this image:

   pixel(point) = {
     d = distance(point, center);
     if d between radius1 and radius2
       return white
     else if d between radius2 and radius3
       return black
     else
       return transparent // Yes, some parts of the image are transparent.
     end
   }

with some constants center, radius1, radius2, radius3. 

This image will look like a white ring directly inside a black ring on a 
transparent background.

Now comes the actual proof step:

 * Compare these two images (the processed input image and the circle pattern) 
pixel-by-pixel, ignoring any transparent pixels in the pattern image
 * Sum up these differences and divide them by the number of non-transparent 
pixels in the pattern image
 * The resulting value is x%, with x higher than an agreed-on value (e.g. 95%)
 * Thus there is a circle in the image at <center> with radius <radius2>


This kind of proof can easily be verified by a computer program.

"This kind of proof" here means:

-You keep the structure of the proof exactly identical to the example above. 
Some of the constants in it may change.

-The constants used in the pattern creation script must pass basic plausibility 
tests and possibly some heuristic ones (e.g. regarding the rings' thickness, 
i.e. radius2-radius1 and radius3-radius2).

-The image preprocessing step does not really have to be checked unless you 
replace it with a more complex computation that can actually add structure to 
an image.


Has this been done before? Proofs as part of image recognition?

Cheers
------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8ca0db34fc106b16-M1d5302d5f7796004555ee222
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to