How to Redact Faces and Why It’s important to Get It Right

Your face is how people identify you; it would be unfair not to have a say in where it will show up. Imagine taking a jog at Central Park; per usual, you wave and speak to friendly strangers that pass by. A couple of days later, you find out there was a crime committed at Central Park at the same time that you were jogging. Authorities question you about your whereabouts before and after being at the park; you are clearly a suspect.

They were able to use surveillance cameras from different establishments around the park to coordinate and piece together what happened. It was concluded that you did not commit the crime, but in several of those videos, you’re seen waving and having small talk with the criminal. You’re questioned by authorities again about your relationship with this criminal, and you insist it was a simple exchange of pleasantries. You explain it away.

Unfortunately, the videos make their way to court. Eventually, it’s all over the news, and finally, it reaches your peers. You know you’re innocent, and you were able to have a conversation with the authorities to explain the situation, but the people that saw that video without that knowledge, have made assumptions about you and your association with a horrendous crime. What do you do?

In this case, not much can be done to save the situation. In today’s world, where anything can go viral, there is no way you will reach the same audience that saw the videos. Even if you did, not everyone would be convinced that you had no association with the crime. Your life is going to change.

Ok, imagination over; that will never happen to you. Even if you were in the same scenario, you would be protected by the use of facial redaction and the many laws that require it. This scenario might feel far-fetched, but it has become an everyday possibility.

There is no knowing when you’ll be an innocent bystander or accidentally put yourself in a questionable position. You should have the right to remove your face from the situation and continue your regular life.

What It Means To Redact

“Redact” may not be a word you hear daily, but its meaning is pretty simple. The Oxford dictionary defines it as “to censor or obscure for legal or security purposes”, which essentially means to hide or remove. While “redact” usually refers to document redaction, there is also facial redacting which means blurring and blocking out faces not to have the person recognized, and audio redacting, the ability to mute or bleep portions of what’s being spoken that may contain private information.

Facial redaction was not popularly used until the early 2000s, which is no surprise since it follows the heightened use of digital cameras and a spike in social media usage. Though it’s hard to say precisely when facial redaction became widespread, the increase in risk factors that correlate with privacy makes the use of facial redaction inevitable.

When you choose to redact faces, you have several choices as to how this can be done. Some popular ways to redact faces are to blur, pixelate or fill them. Every digital image is made up of picture elements called pixels.

To redact, the overall process is focused on replacing those pixels that make up the face with either other pixels that surround it, with different neutral-colored pixels, or with one specific solid color to portray a block covering the face. The goal of redacting images is to secure the person’s identity while allowing the image to maintain its context.

How to Redact Faces

The steps used in redacting are almost the same from one redaction software to the other.

For the manual options, you simply use your mouse to drag across all the faces you would like to redact and choose the effect. Some redaction tools also offer automatic options that allow you to redact faces with a push of a button.

Many redaction software also offers the same effects to redact faces, blurring, pixelating, or blocking the faces. What sets some redaction software apart is the precision of what is redacted and if it is being redacted correctly. This means it should be able to detect various facial features no matter the size, position, resolution, or skin tone and redact.

The detection should also be able to redact faces and faces only. It has to have as few false positives as possible, meaning it does not detect a random object and concludes it’s a face. When a face is being redacted, the system should be able to detect it from different angles, including from the back of the head. This brings along the critical difference between face and head redaction.

Typically, when we say “face redaction” most mean the entire head, but behind the scene, if the AI were only detecting faces, it would miss a lot of its target.

A well-developed redaction software should also carry a more intuitive user interface to allow users to work efficiently with the software, and it should have a processing speed that is fair to the assessment.

This means the users should not have to wonder how to select an effect; it should be evident with either properly conveyed icons or correct labeling. And, with the right equipment, like video cards and storage space, the process should be quick. Redacting faces in a 10-minute video should take 10 minutes or less.

Lastly, the software should be able to redact images of different formats; we have so many file formats that if a redaction software can only work with one or two, it is ultimately useless.

The Good, The Bad, and The Not-So-Worth-It Redaction Software

Let’s compare what good, bad, and “yikes” face redaction software results look like. We’ll use the image below for the demonstration.

You can see several faces and heads in this image, and not all of them are facing forward at the camera. You have multiple heads in the back as well, making it difficult for even the human eye to gauge.

Some faces are turned to the side, and some have the back of their head to the camera. A good AI Automatic redaction tool will be able to detect and redact all of them.

Example 1:

For the first example, I will use CaseGuard Studio.

As you can see CaseGuard AI Automatic Redaction software is easy and intuitive. Here is the final result below. It takes a few clicks to use the AI Automatic tool in CaseGuard Studio. It was able to correctly detect and pixelate all the faces and heads without skipping a beat. WOW!

Example 2:

I’ll now use a different redaction software to compare its results with CaseGuards.
While still using pixelating effect, here is a result of an Automatic face redacting tool that has been used to redact the same image, which did not do an excellent job of redacting all the faces in the same image. It can redact the faces that are relatively easy to detect, but you can easily recognize the faces in the distance that remained unredacted in the result below. Some manual intervention will be necessary to finalize your work.

Example 3:

Finally, with a third redaction tool, I’ll continue to use the same image and effect and see how well it will redact faces and heads. Below are the results, and it provides an example of an Automatic face-redacting tool that should not be used with confidence. Though it can redact the faces in the foreground, it cannot redact the back of heads, side profiles, and most faces off into the background. It also does not focus on redacting hair, which can have users conclude that it only uses face redaction, not head redaction. At this point, a lot of the work will be done using a manual tool to complete the task, making the Automatic feature of this software worthless.

Did You Say Un-Redactable?

Being able to redact faces allows privacy for the well-being of the population. However, it’s important to note that the advancement of technology, on one end, means that the same level of advancement can be made on the other. Let me explain.

Remember the jogging scenario where your life was forever changed due to your face being plastered all over the news? You decided to do something about it; you hired a private investigator!

You want to use your social media platforms to show the people in your life, and hopefully others, that you had nothing to do with the crime committed. The plan is to have the PI collect the same videos that got you here in the first place but also to have him find videos from surrounding businesses that show that you went straight home after your jog. To take it a step further, you want videos that show you did not leave home until the following day, long after the committed crime.

The P.I. comes back after several days, saying he has all the videos you’re asking for, but they’re all redacted. Fortunately, you have been doing your research about AI facial detection and redaction and learned that not only can AI redact faces, but it can also un-redact. Yes, UN-REDACT! And you began to build yourself a machine to help prove your innocence.

Since you have this rare un-redacting software, you can feed the machine your picture and have it learn the measurements and dimensions of your face. Now, within all the videos that the PI was able to obtain, the AI can find and un-redact you just enough so that you can be identified. Phew!

Don’t panic at the mention of AI having the ability to un-redact. To challenge and encourage the enhancement of privacy protection technology, AI is trained with deep machine learning to reverse the work of redaction. It was tested using famous and familiar faces, and though it could not be un-redact to perfection, it made the objects recognizable again. Many factors determine the success rate of the AI being able to un-redact, including the intensity of the redaction and being given what to look for.

The chances of it un-redacting are nearly impossible without providing it some sort of aid. Even though this is technically possible, this type of technology is not widely available, so there is no need to panic. Technology is continuously working to best itself. This capability has inspired those who protect your privacy to dive even deeper to find a permanent solution to your privacy needs.