?Filters like Pikazo?

MosaiCat

Swapping the source and target, you get a Furdinsky.

Too... much... fun... don't... know... how... to... stop...

Daaammnnnnn this is getting somewhere! Amazing work ThreeDee!

Hey Sphinx,

Yes, very similar. I see you used a script. I did it manually with difference blend plus IF math, cumulatively adding up ten (offset and noise distorted) versions of the same source image. It ends up with the part that is closest to the target image from any of the ten variations of the source exactly like your Collage. Blur option on the difference blend to make the pieces smoother.

In fact, if I were to make a source file with a bunch of images of people and animals, I bet it would have some resemblance to Google Deepdream images

What hides in the details?

Nice and well done examples, cool technique ThreeDee

Quote
ThreeDee asked What hides in the details?

miau?

More specifically there seems to be at least two cats, one striped and one orange.

On the first image above it seems to be a woman or maybe more than one

Quote
Sphinx wrote Actually my Collage filter is doing something similar: from the provided sources it uses the one that comes closest to the main input image.

YES, the first thing I thought after seeing the images from ThreeDee was in your Collage filter, and is I have told in the comments thread of that filter is really very well done and clever

Van Garlic

Lady Gogh

Window Gogh

There's probably an optimum for how many copies of the source to use. I've had success with anywhere from 5 to 10 copies. With less copies you tend to see the original too clearly. with more copies you don't see much details from the source image. In my latest version I also rotate each copy randomly, which lessens the likelihood of repetitive features and improves the chances of getting an edge or line feature landing in the right spot.

There's also the weighing matter that can be improved. Meaning, for low contrast areas one could use larger chunks of the source image and smaller chunks for high-contrast areas, high-pass/edge detection style.

---

I initially tried to do "pattern matching" with Bomber+ but it got train wrecked due to not being able to use a blur component. The idea was to pick a number of random pieces from the source image, overlay them onto a same-sized piece the target image given by the Bomber+ corners and use the one that matched the closest as the particle. However, I can't figure out a way to find the closest match without blurring the difference results.

I guess it could be scripted like we've done with the "internal image average". Or does Script component ruin the Bomber+ chain as well? It's not listed as a bitmap-based component, but somehow I suspect it could be trouble regardless.

Had to try it: Picasso by Picasso. (In FF)

Quote
I knew this could be done in FF...it just needed some of the real legends of the community to take a crack at the process!

I don't think it's the same thing, but I get your point now. So in a really really basic sense you're saying that the
feature recognition --> convolve --> apply to target can much be abstracted to suit filterforge. So while the methodology can be different we're still achieving some level of similarity to the original idea?

Interestingly, a search on "collage" (I was looking for Sphinx's apparently now-classic example) seems to produce a lot of filters that might be useful for deepdream and "computer cubism" as a larger discipline.

I wonder if Picasso's spirit is somewhere looking down on all this, and what he would think of it all.

https://www.filterforge.com/filters/se...ollage&h=r

So then you feed the above cat photo as source and then I choose the following image as style

Using the same cat and as style

Using this style with the same first cat

Using this ferrofulid watercolor that you may know from another thread in the forum

Then is curious how much is modified the style sometimes according to the source image, as here is tree and in the result there is nothing related to a tree

same first cat and as style this tree

This an artwork from the great and excellent (for me at least) painter Jackson Pollock

Now using this cubism artwork from the very good artist George Braque

Now using another different source image

And also on the above image result using this style that I have used already and got the faded gray, here is not shown this

Style



As you can see here on this result that all colors are the same