Think you can get lost in the crowd?
Not anymore

This photo was taken in Canada and shows about 700,000 people.

Pick on a small part of the crowd click a couple of times -- wait -- click a few more times and see how clear each individual face willbecome each time. Or use the wheel on your mouse.

This picture was taken with a 70,000 x 30,000 pixel camera (2100 Mega Pixels.) These cameras are not sold to the public and are being installed in strategic locations.

http://www.gigapixel.com/mobile/?id=79995

 

They can 'see' you but can they recognize you in the sea of data generated from these systems. If you understand the algorithms used it's possible to avoid computer vision auto face recognition. If a computer scans the camera image it doesn't see the small section of the large image containing each person as a face.

http://cvdazzle.com/
http://www.computervisiononline.com/blog/creating-art-while-challenging-face-recognition-algorithms
http://rstb.royalsocietypublishing.org/content/royptb/364/1516/529.full.pdf

 

If a computer scans the camera image it doesn't see the small section of the large image containing each person as a face.

Oh c'mon. That could be fixed with a couple lines of code and some processing power. If they want to recognize you, they will.

 

These cameras are not sold to the public and are being installed in strategic locations.

I saw a similar photo taken at the Blackhawks game where they won the Stanley Cup. Almost everyone in the arena was identifiable. It's a brilliant way to sell pictures. One shot, thousands of potential customers.

I noticed there are a quite a few artifacts in that photo, two-headed people and such.

 

Oh c'mon. That could be fixed with a couple lines of code and some processing power. If they want to recognize you, they will.

We both know that computers only do what we program. Programming face recognition in non-perfect conditions is a lot more than a couple lines of code. There is a lot of science being devoted to digital camouflage and systems like CVDAZZLE are just the first stage.

 

Ah, but the majority of those people are not hiding and the photo contains plenty of resolution to identify most of them. Picking out the one guy that's hiding in plain sight, yeah that could be very difficult.

 

Ah, but the majority of those people are not hiding and the photo contains plenty of resolution to identify most of them. Picking out the one guy that's hiding in plain sight, yeah that could be very difficult.

That is a weakness of using mass CV systems to find people who don't want to be 'seen' easily exploit. The point is to ID the person(s) in a large group of people if you need to find someone. These systems generate massive amounts of video or image data that can only be really useful with automated systems to reduce the massive amount of random blocks of image data to faces and then to a subset of similar faces for the humans to match as they are much better than any current machine. The key is to know the weakness in those systems to keep the image from the human operator in the first place and it's not really that hard to do.

Most modern image CV systems now use some version of sparse encoding that's similar once the system thinks it has a face.
http://www.cmlab.csie.ntu.edu.tw/~yanying/paper/msp022-chen.pdf

 

Facial recognition won't get me!, Not with my tin foil hat on!
(it's a tin foil spartan helmet)

 

Facial recognition won't get me!, Not with my tin foil hat on!
(it's a tin foil spartan helmet)

Facial recognition won't get you but the men in white suits will.

 

Did anyone find Waldo?

 

Waldo no but I finally found Carmen Sandiego! !

 

Waldo no but I finally found Carmen Sandiego! !

Where in the world was Carmen Sandiego?

 

https://ahprojects.com/projects/hyperface/

HyperFace is a new kind of camouflage that aims to reduce the confidence score of facial detection and recognition by providing false faces that distract computer vision algorithms. HyperFace development began in 2013 and was first presented at 33c3 in Hamburg, Germany. HyperFace will launch as a textile print at Sundance Film Festival on January 16.

HyperFace works by providing maximally activated false faces based on ideal algorithmic representations of a human face. These maximal activations are targeted for specific algorithms. The prototype above is specific to OpenCV’s default frontalface profile. Other patterns target convolutional nueral networks and HoG/SVM detectors. The technical concept is an extension of earlier work on CV Dazzle. The difference between the two projects is that HyperFace aims to alter the surrounding area (ground) while CV Dazzle targets the facial area (figure). In camouflage, the objective is often to minimize the difference between figure and ground. HyperFace reduces the confidence score of the true face while shifting confidence to the nearby false face regions. This project opens new opportunities for altering the environment and architecture to impact and undermine the effectiveness of computer vision.

Conceptually, HyperFace recognizes that completely concealing a face to facial detection algorithms remains a tehnical and aesthetic challenge. Instead of seeking computer vision anonymity through minimizing the confidence score of a true face (i.e. CV Dazzle), HyperFace offers a higher confidence score for a nearby false face by exploiting a common algorithmic preference for the highest confidence facial region. In other words, if a computer vision algorithm is expecting a face, give it what it wants.

 

https://ahprojects.com/projects/hyperface/

Interesting cat and mouse game. I think the net result is that the cat will get better faster. I'm not sure that's a good thing.

 

Interesting cat and mouse game. I think the net result is that the cat will get better faster. I'm not sure that's a good thing.

CV Dazzle (to obscure the real face) and HyperFace (to provide a facial target) when combined should offer pretty good protection from casual group facial recon systems. Sure the cat (computer vision system) will get better but counter-countermeasures have a way of making surveillance systems more inefficient over time.

 

I wish I had the link to share. Something we make here goes into some drone and test footage has been posted. The camera involved is some mash up of hundreds of cameras. Think of 100 cell phones side by side.

The imagery shown was of a small city. You could barely see traffic moving till the scene was zoomed in. Then you saw cars, people, the works. The narrator then chose random people to click on and the path they had taken was highlighted.

Works in real time once you have enough history, or just archive it and analyze to see what anyone does all day.

 

 

That is a weakness of using mass CV systems to find people who don't want to be 'seen' easily exploit. The point is to ID the person(s) in a large group of people if you need to find someone. These systems generate massive amounts of video or image data that can only be really useful with automated systems to reduce the massive amount of random blocks of image data to faces and then to a subset of similar faces for the humans to match as they are much better than any current machine. The key is to know the weakness in those systems to keep the image from the human operator in the first place and it's not really that hard to do.

Most modern image CV systems now use some version of sparse encoding that's similar once the system thinks it has a face.
http://www.cmlab.csie.ntu.edu.tw/~yanying/paper/msp022-chen.pdf

While I agree that it is currently pretty easy to evade facial recognition systems (and, perhaps, always will be) there is still the human factor. How many humans actually KNOW how to do so and how many of THEM actually DO do so, and how many of THOSE actually do it religiously every where they go.

Another point is that, regardless of the current state of the art in CV systems, they are worthless without the data they need. Cameras like this are tackling that problem.

Personally I have very mixed feeling about this technology. On the one hand it has significant potential to locate criminals and, possibly, terrorists and the like. On the other hand it opens up the door (or opens it wider) for the government tracking the movements of average citizens. That really concerns me.

 

While I agree that it is currently pretty easy to evade facial recognition systems (and, perhaps, always will be) there is still the human factor. How many humans actually KNOW how to do so and how many of THEM actually DO do so, and how many of THOSE actually do it religiously every where they go.

The people that will know how to evade these types of systems are potentially the same people the systems are designed to find in an emergency but it's usually after the fact. Almost all these system are really used in reconstruction of events not prevention. We usually will be able to ID them after the fact but I don't think that video technology will every stay ahead of simplistic countermeasures in real-time unless we have prior knowledge of possible events and locations..

 

just archive it and analyze to see what anyone does all day.

This was invented by folks at USAF and Haas been weaponized.

Essentially a drones would be surveying a large areas archiving lots of video footage. Once an IED has exploded, they would analyze the footage automatically and track those people back to their homes.

Pretty cool way of fighting crimes.