Making masks -- warning double nerd alert

[rutt]

Warning, this is only for the most serious kind of nerds.

Does anyone know this work: http://grail.cs.washington.edu/projects/digital-matting/image-matting/

I've seen the results in the paper and other results in a digital photography course I'm auditing at MIT (http://groups.csail.mit.edu/graphics/classes/CompPhoto06/). Amazingly good.

Now I want the thing. It solves what I consider to be the absolute most frustrating, time consuming problem in image post processing. I've been known to oursource. I've used Knockout. This technique could power a Knockout clone that would actually work.

Anyone want to whip up a Photoshop plugin overnight?

[jfriend]

Now I want the thing. It solves what I consider to be the absolute most frustrating, time consuming problem in image post processing. I've been known to oursource. I've used Knockout. This technique could power a Knockout clone that would actually work.

Anyone want to whip up a Photoshop plugin overnight?
If it's as good as it says and somebody integrated it into CS2, I'd buy it too.

[DoctorIt]

So as Rutt described it to me, this is image processing done by super nerds in environments such as Matlab because "photoshop can't handle it". If anyone here figures out how to implement this, they will definitely be crowned uber nerd of dgrin!

[BrettG]

<snip from site>
Addendum
We forgot to mention one thing in the paper... We used eigen-analysis to find the orientation of the covariance matrix and added \sigmac2 in every axis. That is, we decomposed \SigmaF as U S VT. Let S=diag{s12,s22,s32}, we set S'=diag(s12+\sigmac2, s22+\sigmac2, s32+\sigmac2) and assign the new \Sigma_F as U S' VT. ...


Oh well then... over sigmac2... well sure...

:bash

(I got nothin)

[Mike Lane]

<snip from site>
Addendum
We forgot to mention one thing in the paper... We used eigen-analysis to find the orientation of the covariance matrix and added \sigmac2 in every axis. That is, we decomposed \SigmaF as U S VT. Let S=diag{s12,s22,s32}, we set S'=diag(s12+\sigmac2, s22+\sigmac2, s32+\sigmac2) and assign the new \Sigma_F as U S' VT. ...


Oh well then... over sigmac2... well sure...



OMG that's, like, so obvious LOLZ!!!1!11!!!!

Just the other day I was like d00d, we totally decomposed \SigmaF as U S VT. And d00d was like no duh, you used eigen-analysis to find the orientation o the covarience matrix and added \sigmac2 in every axis.

Rad. c u l8r y0!

[rutt]

If it was easy, everyone (err, someone) would have done it.