ID:34559
 
Hat tip to Little Green Footballs, here's an excellent video on an interesting new technology in image resizing.

http://www.youtube.com/watch?v=c-SSu3tJ3ns

The method is called re-targeting, in which an image isn't cropped or scaled, but columns or rows are removed from less interesting parts of the image. What's cool about this is that it minimizes distortion by using "seams", staggered lightning-like lines, instead of straight columns or rows, so it can snake around desirable features.

Among other cool things this can do are protect faces from distortion, and entirely eliminate unwanted features in an image.
All I see is a white box =/
Images are now even less evidence than they used to be.
I saw the video before, but it's really amazing how you can resize, and actually get the original image back.
Wow. That is pretty nifty. I could see that being used in a number of places, but it appears implementation may be difficult. There are four or five algorithms to choose from, depending on what your image is like.
That is amazingly incredibly stupendously awesome. :-O

I look forward to the day that this kind of tech is built into image authoring tools and web browsers!

@Yota: That's not so bad; the basic algorithm is the same, it's just the importance map that gets changed, and those look like fairly straightforward functions to implement.

@FS: Check your Flash install.
Very nice! There's a part of me that doesn't like the fact that it literally changes the image composition. For example, where an original image contains 60% clear sky, the resized image will contain less.

The practical portion of my brain tells me that it's no big loss, since the space removed by this method will most often be "empty" space, but the more artistic side of my brain is wincing.

Of course, that's no big deal, anyways. For those images that would truly lose something if resized by this method there's still always the traditional resizing algorhithms.

And to add on to Crispy's reassurance to Yota:

Current image editors have multiple algorhithms to choose from when resizing (bilinear resampling, bicubic resampling, pixel resizing, "Smart Sizing", etc.) And people have already been choosing from them to find the best results (Resize, Undo, try another method). For instance, when resizing pixel art, it's always best to use the direct pixel resize to preserve the layout of the original pixels, while the other methods work better for higher-res images.

This would just add some more options to the drop down selection.
I'm pretty sure you need a .com there rather than .org, unless the news came from a domain-parked link farm thing.
I was interested when I saw this, but I would need to see it in action to believe it did a half decent job. You really can't make out the seams from the low res flash video, but I bet many of those images had really ugly seams in them.
Saw this on digg a while ago, good stuff! I thought it would be usable in web design, until they started adding in pixels to stretch it farther than the original.

To me, that just might be a little much (but still cool.)
Yeah, it could work for some webdesign, but I normally just simulate the same thing with a few background images, tables, and some crafty HTML work. Of course, this only works on simple layouts, and having it done automatically would be great.
Bah, they obviously stole this idea from my MapCompress library. :)

Kidding aside, the video is pretty slick!