The many dangers of Facial Detection and Recognition Technology in the Digital Age

Author: Avi Parshan

This was originally formatted for a presentation at my college, but I found the topic interesting and extremely relevant. Therefore, I made the decision to publish it on medium and my blog.

You will learn that facial recognition is advancing at a scary pace and its implications will have a lasting effect on all of us.

What makes your face unique from anyone else’s?

Here is a simplification of this long process which I broke down to 3 steps:

A popular method of detection

· is the Viola-Jones Algorithm (Paul Viola and Michael Jones), invented way back in 2001 and I’d like to expand on how it works.

o Changes the image to monochrome

§ Then it splits up the face into 4 regions

· Your face has different landmarks… regions around your cheeks are generally much lighter than the regions around your eyes – referred to as “Haar Features”

· It can infer the difference between the dark and light points and match it within it’s algorithm

o Now, it can determine if there is a face in an image with surprisingly good accuracy. It then draws a bounding box around the face and crops around it

Let’s move on to facial analysis:

· We use Facial Geometry:

o Which measures the distance between your eyes, nose, and bridge

o In addition to several more points such as the depth of your eye sockets

Lastly, we have the Signature stage:

· Convert the facial geometry mesh into data – or a “face-print”

· This information is then uploaded to a database and is compared with other face-prints to find a match.

Technical Innovations:

· Why has facial recognition become more prevalent?

o HD security cameras can zoom in and can pick out individuals in a huge crowd while maintaining quality

§ And more importantly Moore’s law

o Computers are advancing exponentially… which allows us to give complex problems to computers and they can solve them twice as fast as previously.

§ This ties in with Deep Learning (a machine learning technique) which has become popular – for facial recognition among other uses

· To put it simply, researchers give the program a lot of test photos and have it teach itself to match faces on it’s own based on their labels.

Let me elaborate further: on Technical challenges: In the field of Computer Vision, it is nearly impossible to achieve 100% accuracy at any given task. But it makes educated guesses and marks them with a “confidence” rating. This can lead to false results.

One company developed an algorithm to recognize individuals based on their skin pore location and sizes. In other words, even twins can be detected as 2 separate people.

Next, using a picture or wearing a mask of a victim’s face. Some engineers developed a liveness test, which can tell if a person blinks or twitches to prove that they are alive.

Relevancy:

We have to remember that, the tech exists in all of your smartphones and computers.

· Auto-Focus

· Background replace tool on your Zoom.

· Face-ID to unlock your phone

· Face filters on social media apps

o Age yourself, add facial hair, and even change genders

· Google Photos … it can find certain people and then group their pictures together.

But at the same time, it has many downsides:

· Obstructions

§ Different lighting issues can confuse the algorithm and prevent it from recognizing people’s faces

§ What if you aren’t looking directly at the camera

· Some algorithms project peoples face onto different objects

· While other ones grouped several photos of the same person together to solve this issue

§ As you’re aware people wear Face masks – due to COVID

· The camera will only be able to see half of your face

o So how does it overcome this major roadblock?

§ Each algorithm is implemented differently but the consensus is that they need to base the recognition on upper facial features.

· such as distance between your Eyes and eyebrows.

Concerns

§ Constant surveillance

· This is important to me because:

· Our faces are being stored in government or private databases and Unlike fingerprints, facial recognition can be taken at a distance without the person knowing

o Track our location, know of our whereabouts, who we meet…

· Can’t exercise first-amendment rights

o Recognizing protestors and then link them up with social media profiles and targeting them for potential arrests.

· Racial and Gender Bias

o According to the study “Gender Shades” a MIT project

§ Which tested programs of companies such as Google, IBM, and Microsoft.

§ According to their research: The error gap between lighter males and darker females is as much as 36%

· In other words, people of different races and genders are usually neglected by the algorithm.

§ Data is trained mostly on white males – “it’s not facial recognition but rather racial recognition/discrimination”

· As the pace of innovation increases, facial recognition will also improve, and its implications will have an everlasting effect on us all.

· Whether we like it or not, this technology is here to stay.

Sources: