?

Log in

No account? Create an account
entries friends calendar profile My Website Previous Previous Next Next
"Seam Carving for Content Aware Image Resizing" - Mark Atwood
fallenpegasus
fallenpegasus
"Seam Carving for Content Aware Image Resizing"
I remember the first time I saw an animated GIF on a web page. I knew what an animated GIF was, because I had read the GIF spec years before that, but you never really encountered them. Now they've moved from a weird exotic to a standard and often tacky part of everyday experence.

I just saw something today that's amazing, brilliant, cool, and obvious in retrospect (it could have been in the net-pbm tools 10+ years ago, if anyone had thought of it), that is currently research demo, but I predict that in a couple of years, it will be an unremarkable and standard part of the user interface, and UIs that don't have it will seem stupid and clunky.

It's called Seam Carving for Content Aware Image Resizing, and it's a way to resize and rescale images in a way that's useful, fast, headslappingly obvious in retrospect.

Tags: , ,
Current Location: Home, Capitol Hill, Seattle WA

8 comments or Leave a comment
Comments
hollyking From: hollyking Date: October 2nd, 2007 06:44 pm (UTC) (Link)
Interesting.
jatg From: jatg Date: October 2nd, 2007 06:48 pm (UTC) (Link)
Wow...that's more than a little amazing!
elfs From: elfs Date: October 2nd, 2007 07:48 pm (UTC) (Link)
Liquid Rescale for Gimp, a free, open source implementation of the algorithm described in the paper by Shai Avidan and Ariel Shamir:

http://liquidrescale.wikidot.com/
From: technoshaman Date: October 4th, 2007 03:38 am (UTC) (Link)
Cool and double cool. Not just a really slick algorithm, but a Free implementation of it.

Wonder how CPU-intensive it is?
fallenpegasus From: fallenpegasus Date: October 4th, 2007 05:12 pm (UTC) (Link)
It appears to be not to terribly CPU-intensive, especially since you can precompute and store the energy function and the h-order and v-order pixel ordering, and once you have that, removing lines can be done with a mere offset lookup.

Storing those 3 values per pixel will roughly double the size of the stored image, which isn't too horrible, and would be very very easy to as a set of TIFF planes, or as extention data in JPG, GIF, and PNG. I wonder what file format his demo is using.

My two big questions are:

One, how well does it work with video? Would the background and relative seperations of objects "crawl" back and forth in a moving video where this transformation had been done? Maybe the solution to that is that the energy function has to have a temporal element as well, and reach forward and backware in time.

Two, how well does it work with JPEG block artifacts. Said artifacts are usually nearly invisible to the unaided untrained human eye, but they would be clearly visible to the energy function calculation.
zonereyrie From: zonereyrie Date: October 2nd, 2007 07:49 pm (UTC) (Link)
Yeah, I posted about that a while back, it is very cool. And it does seem obvious once you see it - but lots of great ideas are like that.
intrepid_reason From: intrepid_reason Date: October 3rd, 2007 01:43 am (UTC) (Link)
I have always done something similar when working with images...but this was slick! I don't have photoshop so I'm not sure if all those functions are available now, I settle for using the free GIMP.
captain_button From: captain_button Date: October 3rd, 2007 01:16 pm (UTC) (Link)
"Now they've moved from a weird exotic to a standard and often tacky part of everyday experence."

You mean like web pages that tell you nothing useful if you have active content turned off, unless you download a 20 megabyte .pdf file?

My irony sense is tingling....
8 comments or Leave a comment