Dear Lazyweb: Looking for a tile-fitting program

[elfs]

Dear Lazyweb:

I'm looking for a tile-fitting program (or even library) that can assemble several photos of different sizes into a single, rectangular mosaic. I does not have to be (I would prefer it not to be) one of those "photomosaic" things where it transforms the pool of images into pseudopixels for a parent image; it may downsize the photographs, but must have a configurable minimum size; it may not change the ratio of the photographs, nor rotate them. Basically, I want a program that will assemble wallpaper out of a photograph collection.

Is there such a program? I've spent hours looking for such a thing without success.

Comments

[lucky-otter.livejournal.com]

If you'll accept blending of edges, http://research.microsoft.com/en-us/projects/i3l/autocollage.aspx

[elfs.livejournal.com]

Thanks! The program's not useful to me (blending is unacceptable, and I can't run MS software anyway) but the paper that came with it, "Efficient Optimization of Photo Collage," is awesome bedtime reading!

[lovingboth]

This must be doable with ImageMagick.

[elfs.livejournal.com]

The tiling algorithm is far beyond anything in the ImageMagick montage suite. Believe me, I've looked.

[lovingboth]

Because doing a 'resize' on the images and then 'convert -mosaic' the results will leave gaps if you're mixing landscape and portrait aspect photos?

Can you post some example images of what you'd like to do?

This isn't a panoroma or an overlapping collage, correct?

[pakraticus.livejournal.com]

If it was a panorama, I would suggest autopano-sift.

But this problem space almost sounds like shape fitting for chip layout.
Which when I tried searching for "rectangle fitting" coughs up the terms "2d bin fitting" "2d cut stocking"

If you use your minimum reduced size as the quantum there appear to be two interesting solutions within http://stackoverflow.com/questions/143580/where-can-i-find-open-source-2d-bin-packing-algorithms

I would suggest having some quantum that one or both dimensions of a resized image must be a multiple of, and have your minimum image size a small multiple of that quantum. If you want a tight fit, also permit cropping.

Now you can go looking for imaging analysis software to find which images will lose the least information when reduced in size :-)