I tried Amazon’s controversial facial recognition software

I tried Amazon’s controversial facial recognition software

Posted by


What you’re seeing right now is what my face
probably looks like to a facial recognition software. It detects common features in a face, keeps
track of them and creates a unique profile. You probably know this technology already,
it’s used on social media a lot to identify people in photos. But facial recognition is also used by law
enforcement. Amazon has been criticized recently for selling
their recognition software to police departments because many civil rights organizations like
the ACLU believe this could easily lead to violations of civil and human rights. They fear police could use this system not
only to track criminals but also to identify innocent citizens. Well, because anyone can use Amazon’s software,
I decided to see for myself what the software can do and how it works. Big IT companies have been working with governments,
military and police for years but after Amazon was recently criticized by shareholders for
their visual analysis software “Rekognition” there has been a new controversy about tech
companies and law enforcement. “Rekognition” is part of Amazon’s Web Services
and it can analyze images, videos and live streams to find objects, scenes and faces. It then generates labels based on the image
content along with a confidence score. The score is a number from 0 to 100 that indicates
the probability that a given prediction is correct. The labels can be for example people, food,
animals or vehicles. This is the web interface of Amazon’s Rekognition
and it’s possible to do a demo right on the website. Once a photo is uploaded, the software will
analyze the contents and produce the labels. Normally these results would be machine readable
code but this demo also generates human readable results. Although it didn’t work for every image I
tried, this detection was quite accurate. Most of these labels seem to describe the
scene correctly. The facial analysis feature returns attributes
like gender, age, emotions, beard growth and glasses. It can analyze up to a hundred faces in one
image. First I tried the facial analysis with a stock
photo of a family and it worked correctly. Rekognition can tell emotions like happy and
sad and the software even detected the babies age and gender, which seems pretty impressive. I also tried the facial analysis with a LEGO
crowd and it actually detected faces. It even analyzed attributes like glasses correctly
although it had trouble with the genders of the figures. Face comparison and facial recognition are
probably more interesting to law enforcement. Face comparison measures similarity by comparing
one face to up to 15 faces in one target image. Facial recognition on the other hand is able
to perform a real time search against collections with tens of millions of faces. For the face comparison feature, I cropped
out the head of a woman and told the software to look for this person in a similar photo. This example would still be possible for a
human because of the small crowd size but the algorithm really pays off when comparing
millions of photos. It found the woman, even though her head was
at a completely different angle. The result are the precise X and Y coordinates
in the image where the person was found. Most of this is not only possible with photos
but also with videos and live streams. It’s obvious that this is mostly intended
for surveillance footage. Video Analytics is able to track people within
and across shots even when the camera is in motion. The software returns a bounding box and the
face, with timestamps whenever a specific face is in frame. Again, this works in real time and can search
collections with tens of millions of faces and by explicitly highlighting this fact,
Amazon seems to be actively marketing to law enforcement. To demonstrate the video analysis capabilites,
I uploaded stock footage of people walking by. The software returns the faces it recognized
and where in the video they are visible, indicated by the yellow bar above the player. Amazon Rekognition can also recognize text
in images and detect adult content. The capabilities for automated mass video
surveillance certainly exist so it comes down to a moral standpoint if Amazon continues
to provide governments with this technology. I doubt the protest of a small portion of
shareholders will be enough to change Amazon’s mind because there is a huge potential for
profits. And Amazon is by far not the only one working
like this, Microsoft for example worked together with the US immigration enforcement to manage
and analyze data about the identity of people in real time. Google is trying to end a project with the
US Defense Department that helped recognizing objects from drone videos. Recognition technologies by themselves aren’t
the problem, in fact they are definitely needed. The amount of video content websites like
YouTube already have to deal with today is far more than any reasonable amount of human
workers could filter. But the relationship between technology companies,
law enforcement and military are in my opinion a valid cause for concern, especially because
this market is only going to get bigger with more precise algorithms and too few people
seem to care.

15 comments

  1. “NEW! CONTROVERSIAL Software
    Exploits $12.3 MILLION LOOPHOLE
    & Makes Us $519/DAY…”

    “Find Out How You Can Start Today In Under 5 MINUTES & 17 CLICKS…”
    Get Started In Your Country with Your FREE Software License Here www.lendingclub5k.com

  2. If you are not doing anything wrong, you shouldn’t have anything to worry about. Blame terrorists and people trying to hide their identity.

  3. Thank you for the content, very professional and English is very well spoken, thanks again from abroad; ps ( don't concern yourself with negative comments from the U.S; there are many arrogant people here and selfish, sorry; best wishes, Stephen. 🙂

Leave a Reply

Your email address will not be published. Required fields are marked *