Transcription
In my last video, I discussed how in 2019, Kroger and Microsoft issued a joint press release announcing their partnership to quote, "Redefine the customer experience at Kroger grocery stores." Part of their plan involved using facial recognition to monitor customers as they move throughout the store. And we are talking about hundreds, if not thousands of cameras per store, as they would be built into the digital price tags on every shelf.
But Kroger was not alone. A 2021 article in Vox discussed how retailers from Macy's to Albertson's are also using facial recognition, often without the customers knowledge or consent. And not to be outdone on the creepiness scale, in 2017, Walmart filed a patent application on their plan to not just use cameras to recognize customers, but to read their facial expressions in an attempt to read their minds regarding customer satisfaction.
As always, links to all my sources can be found in the description of this video. The use of facial recognition is not limited to just retailers. It is also being used at concerts and sporting events, by social media apps, on public transportation, and in office buildings, hotels, and gyms. If you drive a new vehicle, your face may even be scanned every time you get behind the wheel.
This is the first in a short series of videos on how to prevent facial recognition from tracking you. In this video, I will briefly discuss how facial recognition works at a very high level. I will then discuss the difference between normal cameras and infrared cameras, and then I will share part one of my test results with you where I tested ways to defeat facial recognition cameras. I am Dr. John Padfield, an engineer turned state representative turned business professor, and this is Business Reform, where we discuss issues at the intersection of business, technology, and society.
To keep the YouTube algorithm happy, I want to explicitly state this video is intended for people who simply want to enhance their privacy and is not intended to enable or encourage any illegal activity.
The first step in the facial recognition process is capturing an image. The image could be from the front-facing camera on a smartphone or a live video feed, such as a security camera in an office building or airport. There really isn't anything legal you can do to prevent a camera from capturing your image, especially if you don't even know you are on camera.
The second step is for the algorithm to detect whether there is a face in the image that was captured. While this sounds easy, sometimes algorithms make mistakes and see faces where none exist, or they may overlook a face that really does exist. This process step creates a potential vulnerability in facial recognition systems known as adversarial images. There have been numerous articles published in academic journals about adversarial images, and we will discuss them more in part two of this video series.
If the algorithm detects a face, it moves to step three, which is called feature extraction. In this step, the software maps the image of the face by finding points such as the edges of a person's jawline, lips, nose, and especially their eyes. Exactly how the software does this is beyond the scope of this video. Once these points have been identified, the algorithm conducts a series of measurements between those points, such as comparing the ratio of the width of an eye to the distance between the eyes, the width of the nose to the length of the nose, and the width of the mouth to the distance between the mouth and nose, etc. Once all those ratios have been calculated, the face can be described by a series of numbers representing those ratios. While the actual process is more complex, this gives you a general idea of how facial recognition works, and it is accurate enough for the purpose of this video.
The feature extraction step is where we have our best opportunity to defeat facial recognition. If we can interfere with the algorithm's ability to extract features around our eyes, it will severely limit the system's ability to describe a face via a series of measurement ratios, as it will be unable to calculate several key ratios.
Assuming feature extraction was successful in step three, the algorithm will then take that series of facial features from a captured image and compare it against known faces, looking for a match. And finally, in step five, the system will produce an output that says it either found a match or did not find a match. Or, to be more precise, the system will probably tell you the probability that it found a match, such as a 97.8% chance the image captured was a particular person.
Some systems, such as facial recognition on a smartphone, perform facial recognition verification. This means it looks at a live image and compares it to one stored image and answers the question, "Is the person looking at the camera right now the owner of this phone?" Other systems, such as those used by law enforcement, perform facial recognition identification, where the captured image is compared to millions of stored images to determine who the person in the image is.
If you think you can defeat facial recognition by wearing a wig, a fake beard, or wearing dark sunglasses, I've got some bad news for you. Those techniques simply do not work on most facial recognition systems for one simple reason: infrared cameras.
Everything from AM/FM radios, microwave ovens, and wireless routers to infrared heaters, visible sunlight, and X-rays. They all work on different frequencies of electromagnetic radiation. While we can't see infrared light, those wavelengths are just a little bit longer than red light, and they can easily be picked up by cameras. However, we normally don't want our cameras to pick up infrared light. So, cell phones, GoPros, Canon, and Nikon cameras, they all come with IR filters built into the sensor module. Exceptions to this rule include night vision trail cameras, like those used by hunters to take pictures of wildlife in the woods at night, and security cameras that capture clear images in near total darkness.
Many facial recognition systems use infrared light to capture images. Just like an X-ray can look through your body to see your bones, infrared light can look straight through your dark sunglasses and see your eyes. I am wearing the same pair of dark sunglasses in both of these photos. The photo on the left was taken with my iPhone, which has an IR filter in the camera sensor module. These glasses are so dark that you cannot see my eyes. The photo on the right was taken with an inexpensive security camera using infrared LEDs that can look right through the dark lenses and see my eyes.
But here is a better test. What happens when I look at my iPhone while wearing regular sunglasses? While the iPhone has an IR sensor on its regular cameras, it does not have an IR filter on the Face ID camera. iPhones with Face ID have an infrared projector and an infrared camera, so they can look at your face in total darkness and not blind you with visible light if you pick your phone up in the middle of the night. Thanks to the IR projector and camera, my iPhone unlocks without any trouble at all while I'm wearing my dark sunglasses.
What about blue light blocking glasses, such as this $16 pair from Amazon? The same thing happens. Infrared light passes right through those glasses, and my eyes are easily seen by the camera and unlocks my phone.
However, you can buy IR blocking safety glasses on Amazon for around $13. These glasses are very dark and designed for people who work with certain types of welding equipment. This particular pair has an IR filter shade 5.0 lens, which means in addition to blocking nearly all infrared light, it also blocks 96 to 98% of visible light. While these IR blocking safety glasses prevent your eyes from being seen by an infrared camera and will prevent your iPhone from unlocking, they are too dark to wear for driving and they would not be safe or practical to wear indoors when you go shopping.
Amazon also carries IR safety glasses with IR filter shade 3.0 lenses for around $13. These glasses also prevent your eyes from being seen by an infrared camera and will prevent your iPhone from unlocking. The IR filter shade 3.0 lenses block nearly all infrared light and they block 86 to 90% of visible light, so they are very dark, but perfectly fine for working in the yard on a sunny day.
While the $13 IR blocking sunglasses from Amazon will prevent your eyes from being seen by an infrared camera, they are not the most stylish. And even the IR filter shade 3.0 lenses may be darker than what you want to wear going to a store or going inside of a concert or a sports stadium. This is where a company called Reflecticles comes in. They make a wide variety of more stylish IR blocking glasses. Some of which have a lighter IR blocking lens more appropriate for wearing indoors, and some even have IR reflective frames for an extra layer of privacy protection.
This is not a sponsored video. I purchased two pair of IR blocking glasses from Reflecticles at full price. I bought the IR pair originals with IR dark lenses, and my infrared camera was unable to see my eyes. I also bought a pair of IR cloak glasses with IR light lenses, which lets more visible light through. In the photo on the left, you can see my eyes under visible light, but in the photo on the right, you cannot see my eyes with the infrared camera. Both pair of Reflecticles glasses prevented my iPhone from unlocking. I personally like the Reflecticles IR light lenses because they are light enough I feel comfortable driving with them on and wearing them inside of a store.
While I was reading about a more expensive model of Reflecticles glasses that have IR reflective frames, I had an idea. I am a cyclist and I enjoy riding my bike at night. A few years ago, I bought a lightweight, super reflective jacket to wear on night rides. Under normal light, the material has a dull gray appearance. However, when a camera flash or car headlights hit it, it instantly becomes blindingly bright. I had a friend try it on and I took his photo using the flash on my iPhone, and this is how the photo turned out. Amazon sells hats made of the same material, so I bought one for this video to see if it would prevent my iPhone from unlocking.
As you can see, my entire face, including my eyes, are clearly visible when I am wearing the hat under normal light and under infrared light. So, I really didn't expect this to prevent my phone from unlocking. But to my pleasant surprise, it did prevent my phone from unlocking. I then tried a regular hat for comparison, just to make sure it wasn't just the presence of a hat, and my iPhone had no problem unlocking when I was wearing a regular hat pulled down to the same position on my forehead.
In this video, we have looked at how facial recognition works, how infrared cameras differ from regular cameras, and how we can prevent facial recognition on an iPhone from working by wearing IR blocking lenses or a super reflective hat. I want to stress, just because these techniques work against an iPhone does not guarantee they will work against the facial recognition systems used in stores, concerts, and sporting events. However, these techniques are simple ways you can go about reclaiming some degree of privacy when you're out in public.
In my next video in this series, I am going to do additional testing beyond simply using an iPhone to see how well IR blocking glasses, a reflective hat, and adversarial images work to defeat other facial recognition algorithms, such as OpenCV and Face++. Two popular pieces of software used for identifying people. However, I have a couple of other videos coming out between now and the next video in this series, so be sure to subscribe if you want to make sure you don't miss my upcoming videos.
Have you ever tried any of these methods for defeating facial recognition? Do you plan to try any of them for yourself? Or do you have any suggestions of other techniques that you want me to test? I can't wait to read your comments.
Also, just a quick reminder, I am conducting my Brush Fires of Freedom tour this summer, and I just added a tour stop near Knoxville, Tennessee on July 17th, 2025. Go to brushfiresour.com or click this link for more information.