Awesome new image resizing algorithm

[necroforest]

This was on /., but I figured not everyone reads that and it's cool enough to post on here:

http://www.youtube.com/watch?v=vIFCV2spKtg

[nyeguy]

The worst thing about that is that it is so simple, and it makes perfect sense, so it makes you wonder why you hadn't already thought about it.

But that is really cool. I can see this being really useful in content presentation for the web.

[Amnesiasoft]

Ooh, very nice.

[taggedunion]

That is quite cool and makes the gadget-lover in me squeal with joy.

[Dingbats]

I don't have Flash and I don't want to install the proprietary plugin, so I can't watch it. What is the new algorithm?

[elminster]

oo thats funky, problem is it would require alot of calculations, or if pre-stored would increase size dramatically. Although certianly a nice approach to it, but only useful for a small number of things.

[gmalivuk]

oo thats funky, problem is it would require alot of calculations, or if pre-stored would increase size dramatically. Although certianly a nice approach to it, but only useful for a small number of things.

Shut up. It's awesome, and anyone who says otherwise is itching for a fight.

(I didn't click the link, but I'm assuming this is the one where it finds curves of minimum energy, so you can resize without distorting or losing the edges?)

[skeptical scientist]

oo thats funky, problem is it would require alot of calculations, or if pre-stored would increase size dramatically. Although certianly a nice approach to it, but only useful for a small number of things.

Doesn't seem like it should take that long. Calculating energy at each point is likely a quick process, and only has to be done once, after which it can be updated in the region of a removed line without having to redo the whole thing. Finding a curve which is a global minimum of energy across the image is probably computationally intensive, but finding local minima can also be done quite efficiently. I bet that the resizing the algorithm performs is not significantly slower than the resampling done when zooming in and out on images in photoshop, acrobat, or other similar programs.

Shut up. It's awesome, and anyone who says otherwise is itching for a fight.

(I didn't click the link, but I'm assuming this is the one where it finds curves of minimum energy, so you can resize without distorting or losing the edges?)

Yes. It's obviously limited for many purposes, as image resizing which distorts is frequently not what you want. It does seem like it has some good uses however - the simulation where they cut two people out of a beach scene was pretty cool, and to accomplish the same effect in photoshop would undoubtedly have taken much longer.

[necroforest]

Calculating energy at each point is likely a quick process, and only has to be done once, after which it can be updated in the region of a removed line without having to redo the whole thing.

Well, I wrote a program that does that with 3-dimensional image data; It took about ~10 seconds to do a 2gb dataset. I think it uses the same idea as the algorithm in the video: calculate the gradient vectors using finite central differences then calculate their lengths

[flatluigi]

I really think this is cool.

[infodump] This is Shai Avidan and Ariel Shamir entry into the SIGGRAPH computer graphics conference. A version of the video with better quality (and much larger filesize) can be found here. The paper with the information can be found here (rehosted by me). [/infodump]

Personally, I find this to be a little more cool.