Using Collage Tile Match Script

Cool image. I can't imagine 27.8 hours. On my computer it would probable take a week.

Nicely done, Tran, but it's time for a new PC. lolol

You have more patience than me. I bet you even have a 3D printer. lol

Or a more efficient matching algorithm. 27 hours? For 192 tiles?

The Texan: "my estate is so big, it takes a full day to drive through it"
The Kentucky hillbilly: "yes, I also used to have a car like that."

I don't know of a more efficient algorithm...
It's 192 tiles yes, it also does 8 colors (indexed selections) so there might be more then 1 selection per tile.
And I also offset my tile image by 1/10 of 100 (squared) so it does 100 times the number of searches so that it can find best match.
I have no idea how to make it more efficient.

I am now testing another version that uses random images from unsplash.it for tile instead of one tile image. But for this version i wanted to find even better match by using histogram call which is 5 times as slow instead of the efficient pixelized blur because my selection isn't always rectangle/square. It could be some weird shape so making the rest of the are white or black would not be as accurate as pixelblur makes your r,g,b value into ints... while i think histogram returns real numbers averages which could be more accurate (I am not sure about this real or int) but a histogram of actual selected odd shape area could be more accurate so i thought i'd suffer the wait.

I have shared the script on GL so if people who have super computers can play with it much quicker than i can.

My laptop is a really old piece of crap, i am also running BOINC while it's running so i don't know if that slows it down or not surely doesn't help the wait time also when i try to browse the internet sometimes the mouse stops moving like as if computer was completely frozen then it starts reacting after 10 seconds or so so .... at least it doesn't freeze for real (Thank God).

This edit with not so busy sky. Which one do you like more, busy or not busy sky?

What are the criteria for a match?

Look for the tile (or selected shape which is just the indexed color intersected with tile rectangle) where histogram's sum or rgb of difference mode is closest to black smallest sum
That's it...but i divide the tile image into tiles (while offsetting the whole image tile 100 times [10 horizontally and 10 vertically]).

I used the indexed color and intersect it with tile square to get funky shapes but i still compare to original details of image so that it tries to match with original details (if it can) instead of just the plain indexed colors.

ofnuts,
Have you thought of ways that i could speed up my script?