Hacker News new | past | comments | ask | show | jobs | submit login
OpenBR – An open face recognition/biometry library (openbiometrics.org)
192 points by afics on Aug 5, 2013 | hide | past | favorite | 29 comments

I really like the explanatory graphic on OpenBR's front page. It's certainly not very artistic... But it does a great job of explaining at a single glance what exactly this project does and where it fits in a software pipeline. I wish more projects had an upfront compact visual representation of what they offer.

I can imagine what it would look like if this were a .io startup instead of a research project. The front page would be 20 times taller with endless scrolling, yet consist mostly of beautifully laid out white space filled with some trendy shade of brown. In that vast emptiness there would be floating decapitated heads that actively try to escape the mouse cursor. When I finally manage to hover over one of the heads, it would reveal a nondescript slogan, for example "Recognize. Integrate. Socially." There would be a single cryptic two-line code example given in CoffeeScript, and nowhere on the site would it be apparent whether this product is commercial and how much it costs.

Guys, help me. Every time I see something like this I think "cool thing!", but I also start thinkin about new ways someone will start to track me down.

Seriously, my paranoid level has never been so high.

Not too paranoid, considering the front page of this project says "OpenBR is a collaborative research project started by The MITRE Corporation". What is MITRE?

http://www.mitre.org/work/hs_sedi/ : Homeland Security Systems Engineering and Development Institute—or HS SEDI ...

http://www.mitre.org/work/intel.html : Just about all of MITRE's engineering and science disciplines are involved with improving our sponsors' intelligence, surveillance, and reconnaissance (ISR) activities.

Paranoia can be a symptom of growing or oncoming depression. You should take that seriously: At least take a few days off and try to get a feel for your current mental state.

Studies have shown that depressed people sometimes have a more adequate world view.

Yep, that's called depressive realism. It's pretty bizarre to think about.

It's not paranoia if they're really out to get you.

If it makes you feel any better, their publication[1] reports low accuracy on a popular dataset called LFW (which stands for Labelled Faces in the Wild). Most face recognition technology works very well on mugshots. Frontal Faces -- with almost consistent lighting and alignment. These technologies do not perform well on personal photos.

[1] http://openbiometrics.org/publications/klontz2013open.pdf

You can find OpenBR, along with 40+ other face recognition APIs, libraries, SDKs, and software at http://blog.mashape.com/post/53379410412/list-of-40-face-det...

This is less 'US defence establishment releases software library' than 'US defence establishment requests decentralisation/functional political deniability of widespread facial recognition, offloading of hardware costs to consumer'.

It's actually 'Primary author wanted to spend some time in academia but continue to working on the project'.

Although I currently can't see any evidence of involvement of any government branch, the ultimate result may be the same nevertheless.

Google Glass certainly has the same issue, but to an even higher degree, since we know about the collaboration of Google with the NSA.

EDIT: thanks for the correction, missed that one.

Can someone differentiate between OpenCV and openBR? I think I've read about people using OpenCV for facial recognition. How do these both differ?


The OpenBR API is designed to cater to the researchers and users of biometric algorithms, where as the OpenCV API is designed to solve a more general scope of image processing problems.

OpenCV includes implementations of several seminal face recognition algorithms (PCA, LDA, & LBP). The OpenBR algorithm is more modern/sophisticated in its combination of several well studied approaches. While I haven't explicitly compared OpenCV against OpenBR, the OpenCV algorithms can and have been expressed within the OpenBR and are known to be less accurate. Having said that, accuracy & algorithm choice are highly dependent on image quality. Our technical paper [1] provides more details.

[1] http://openbiometrics.org/publications/klontz2013open.pdf

Thank you, that was very informative :)

OpenCV is about tinkering with your future career. With OpenBR you should open that beer and see where it gets you.

From the OpenBR website:

"Internally the code base uses the CMake build system and requires Qt and OpenCV."

In the unlikely event that you were given a USB License Dongle, then dongle must be in the computer in order to use the SDK. No configuration of the dongle is needed.


Anyone read enough to know when that occurs?

Hi! I'm the primary developer of this project. This paragraph of the documentation is out of date. There is never a need for a dongle, it's all FOSS :)

This is great news! Looking forward to experimenting with this new library. I hope it can be somehow used on Android in the future.

Please feel free to ask questions on our mailing list [1] or IRC channel [2]. Our software dependencies are Qt & OpenCV, both of which compile for Andriod, so OpenBR should be able to run there also, but no one has tried yet.

[1] https://groups.google.com/forum/?fromgroups#!forum/openbr-de...

[2] irc.freenode.net #openbr

I'm curious about how to interpret their 'Face Recognition Evaluation' data though.

I'd like to use this for authentication purposes but if I'm interpreting it right their performance is close to existing solutions, but not close enough for commercial use.

Correct, we do a decent job on mugshots/driver's licenses, but are not at commercial level performance yet. Keep an eye out for NIST FRVT 2012 [1] and our upcoming conference paper [2] if you'd like to see more complete comparisons.

[1] http://www.nist.gov/itl/iad/ig/frvt-2012.cfm

[2] http://openbiometrics.org/publications/klontz2013open.pdf

Nice work! By the way, now that you're here ;) do you perhaps also know if there's a way to address fjcaetano's concerns if you have access to a full video feed?

"Liveness detection" is an active area of research within the community. I know you can enable this check on Android and it works moderately well.

Awesome. Thanks for the reference!

Authentication is problematic, because using a photo can trigger the algorithm to successfully authenticate the user.

I've done this a couple of times with my brother's phone. All I had to do as load his profile picture on my phone and show it to his camera.

Anyone know of a library that can track the direction a user is looking? I don't need exact eye-tracking type stuff from special equipment, just left right up down etc. My initial googles have come up short so far.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact