This isn't like most super-resolution techniques, that take multiple images and align them to a higher-resolution grid. https://en.wikipedia.org/wiki/Super-resolution#Geometrical_o... This is about having a model of what a face should look like, and fitting the data to it. It's like taking a handful of data points and plotting a curve, then interpolating/extrapolating other data points. Except instead of something simple like a bell curve, the model is of a human face! (That means you could feed it random data and it would still come up with a face - grotesquely distorted perhaps, but the model can only make faces.)
The software could've given a clearer image of a suspect which would've aided police in finding said suspect. Figuring out who actually placed the bomb in the first place was bottleneck #1 and probably took the bulk of the time. Bottleneck #2 is finding out who that person is and this is made easier with the crowd-sourcing that is local news. Bottleneck #3 is physically finding the person which they even had trouble doing with Boston on lockdown.
It's a neat solution to a very real problem, so technically very cool just not what the article is billing it as.
Anil Jain from Michigan State University (one of the highest ranked h-index in Computer Science) recently published a technical report on the this subject [0]. The technical reports says that they were able to identify Dzhokhar Tsarnaev with a Rank-1 hit against a 1M mugshot background database.
Except the Rank-1 hit wasn't against an image from the mugshot database.
..The notable rank-one hit for Dzhokhar Tsarnaev is an
illustrative example of this potential. However, the hit was
against a graduation photograph with similar pose that was
tweeted after he had been publically identified, and not a
conventional mugshot from a prior arrest.
Yes but Tsarnaev's brother was known to authorities. It's feasible for authorities to index his photo and known family members.
If they had put both faces into the search, and matched both with high confidence, the result would have sent up a huge flag because the known relationship.
Meaningless unless you know the false positive rate. A technology with a 1% false positive rate over the US population gives you a Minneapolis of people to investigate.
15 comments
[ 3.8 ms ] story [ 40.4 ms ] threadhttp://www.flickr.com/photos/unavoidablegrain/sets/721576288...
The software could've given a clearer image of a suspect which would've aided police in finding said suspect. Figuring out who actually placed the bomb in the first place was bottleneck #1 and probably took the bulk of the time. Bottleneck #2 is finding out who that person is and this is made easier with the crowd-sourcing that is local news. Bottleneck #3 is physically finding the person which they even had trouble doing with Boston on lockdown.
It's a neat solution to a very real problem, so technically very cool just not what the article is billing it as.
That's important because the suspect will most likely flee once they realize you have their image.
Instead you develop leads and set up surveillance.
[0] http://www.cse.msu.edu/rgroups/biometrics/Publications/Face/...
..The notable rank-one hit for Dzhokhar Tsarnaev is an illustrative example of this potential. However, the hit was against a graduation photograph with similar pose that was tweeted after he had been publically identified, and not a conventional mugshot from a prior arrest.
If they had put both faces into the search, and matched both with high confidence, the result would have sent up a huge flag because the known relationship.
They don't want to identify every random person, just persons of interest, wherever they go.