Could you find the missing Hillary computationally?
May 19, 2011 at 9:02 am 4 comments
In our Media Computation classes, we often ask the question, “How could you tell if someone faked a picture in the paper?” Here’s a high-profile example — could you come up with a computational mechanism of recognizing this fake?
A Brooklyn Yiddish-language newspaper airbrushed Secretary of State Hillary Clinton from the White House’s official Osama bin Laden “war room” photograph because editors decided that their Hasidic readership would be offended by a photograph of a woman.
Der Tzeitung ran a copy of the iconic photo of President Obama surrounded by his advisers during last Sunday night’s raid the terror mastermind’s headquarters — but a photo editor removed Clinton from the iconic, historic image because of the paper’s “long-standing editorial policy” to omit women from photos.
via Female problem! Hillary erased from Yiddish paper — because she’s a girl! • The Brooklyn Paper.
Entry filed under: Uncategorized. Tags: Media Computation.
Computing Ed Research - Guzdial's Take 
1.
Max Hailperin | May 19, 2011 at 9:08 am
They also removed a lower-level female staffer whose head is poking through in the back.
2.
Garth | May 19, 2011 at 10:44 am
I guess if you do not like reality, change it. This sort of brings up the whole topic of truth and the internet, is there any relation between truth and the internet? It is very difficult to convince my students that the internet is a terrible resource. I am constantly telling them that they need to be very selective and verify everything they read.
3.
Bart | May 19, 2011 at 11:33 am
Detecting that automatically is a huge challenge and the subject of ongoing research (try “photoshop detect tampering” on http://scholar.google.com) … and then only when having the originally produced digital image.
I think one could reasonably argue that if the tampered image were resampled and say jpeg compressed or printed and rescanned, chances of detecting tampering are close to zero.
4.
Rob St. Amant | May 22, 2011 at 9:35 am
I think it might be a good learning experience, if the students know some basic statistics. They could look at different regions of the altered photo and figure out how to describe the patterns in the pixels (that’s modeling), and maybe find similarities and differences compared with other regions (that’s model comparison). I believe Bart’s observation that this would be very hard to automate for any photo, but I think it would be worthwhile to do for just this photo, to introduce some interesting concepts.