A Face Could Launch a Thousand Tags

Face Recognition Data – should you be concerned? Face recognition has been a hot topic lately with some wide-ranging concerns being expressed about personal privacy. Before we can properly analyze and discuss the potential exposure face recognition poses to our precious privacy, let’s review how face recognition works.

A digital image consists of pixels (dots) of color information. The more pixels in a digital image, the higher the resolution of the image. Camera resolutions are marketed in megapixels. Most cameras today have at least 10 megapixel resolution, which is a lot of pixels. Higher resolution means an image will likely look better. The reason I say “likely” is that a high-resolution image will still, unfortunately, not correct bad photography. Low lighting, poor focus and bad perspective will not be fixed with an expensive camera! Having said that, the same image taken with both a low-resolution and high-resolution camera will appear sharper on screen and in print from the high-resolution camera.

Ok, so we have all these pixels at our fingertips. Now, there are two steps to face recognition. Step one is to actually find a face on the image. This is called face detection. Digital cameras have been using face detection technology for a couple of years. It helps you focus on the people in the photo (as opposed to an item in the background, which can really spoil a nice photo opportunity!). Face detection involves the analysis of the pixels of an image, looking for areas of the image that could potentially contain a face. Some systems look for the eyes, nose and mouth, while others look for skin color. Still other systems combine a number of these factors together to identify areas of the image that could possibly be a face. Current systems have become very accurate and don’t often make mistakes, however there is still going to be an error rate of between 1 and 10 percent depending on the method.

Faces can be identified in regions of the image as small as 20 x 20 pixels, but those faces will be grainy and the subsequent recognition results will not be very good. What this means is that, with a high-resolution photo, a system can detect small faces in the background of the image and actually produce some meaningful recognition results.

Another significant factor in accuracy is whether the person is facing the camera and whether the image contains a full or partial profile of the face. If the system can only see half the face then only half the information is available for recognition purposes. Some systems use a mirror approach to recreate the other half of the face to enable a proper face signature to be created.

Okay, now we have some faces identified within the images. Using various algorithms beyond the scope of this article, the software will develop what is known as the “face signature” for each of the detected faces. The signature can be a single number or a sequence of numbers and is stored in a database, along with a pointer that relates to the specific face region detected within that image. Note that the signatures generated are not perfect. Face pose, lighting conditions, color settings, and focus variations make it a non-perfect science.

The system continues through every face on that image, and then every other digital image that you make available to the system. When completed, the system has this database containing your images, faces and signatures.

Now, step two begins – the face recognition step. Initially, no faces are “known” in your database. The system “knows” a person only when you identify that a specific face and signature belongs to that person. This is accomplished via a process called tagging. The “known” signature can now be compared with all of the other unknown faces in the database. There are mathematical ways to compare the signature of one face to another. These formulas come up with a “likelihood” number that shows how close an unknown face is to a known face. If that “likelihood” number is high enough then the face becomes a suggested match for that person. There is always a “best match” of one face to any other face in the system. It may not be correct, but based on the way the system works and how the data is provided, it is the system’s “best guess”.

How is this information used? Here are just a few examples:

  • With your own digital photo images, the face recognition data is used to help you tag your photos faster.
  • Law enforcement officials use face recognition to quickly match photos of suspects with photos of known criminals.
  • Casinos use face recognition to identify undesirables who enter the casino, and to identify people with gambling problems who are registered with special casino-sponsored support programs.

In the case of your own digital images, if you didn’t use a software tool, the alternative would be to view each photo and attach a name to each face one at a time. While it’s true that your brain does this type of analysis very quickly, the problem with the brain is that the input system (your eyes) is slow when you are tasked with scanning thousands of pictures. After a few hours of this type of work, the typical person requires a rest, change of scenery, or other interruption. And it’s an undeniable fact that “human computer” accuracy falls off in direct relationship to boredom with the task. Needless to say, this is EXTREMELY time consuming and tedious work unless you employ face recognition technology to speed the process.

Once your photos are tagged, your ability to quickly access the photos and share them with friends and family becomes much better. As a result, you can spend more time enjoying your photos versus managing them. Let’s face it, without this sort of tool, very few people ever get around to actively organizing and managing their ever-growing collection of digital photos.

So the $64,000 question is, “should you be concerned if public websites are using face recognition methods to find your face in other photos?”

Face recognition is a valuable tool for your own personal use on your own computer, but we believe there are potential abuses of your privacy if a public website or social network starts collecting and processing face recognition data.

Here’s a quick example. A friend uploads an embarrassing photo of you at a party via a popular social broadcasting service. This photo becomes publicly available on that service, although there is no name attached to the photo. Let’s assume a popular social network you belong to has your face signature data. Another friend sees the photo and decides to upload it to the social network. As the social network scans its system for new photos, it notices this photo and scans it, looking for known faces that are in that friend’s network. Voila! Your face is found and you are automatically tagged in the photo, which may be public on the friend’s social network profile. If so, that photo is now available for the world to see – and it has your name on it.

That scenario is fairly mundane compared to the application of this data for not-so-friendly purposes. In a nutshell, should you be concerned about the application of face recognition data on public networks? When you consider the long-term implication of third-parties “tagging” photos with your name where they are on public display, possibly without your knowledge and out of your control, the answer is a resounding “Yes!”

In order to protect yourself, be aware of all the security and privacy settings on the social networks you participate in, and be diligent about monitoring the public photos containing your mug that are made available to the public via the Interweb. Just as technological advances, such as online banking and shopping have made it easier to conduct day-to-day tasks, so too has face recognition. However, with the increase in online banking and shopping, there has also been a rise in identity theft incidents. Likewise, automated face recognition tagging will generate privacy concerns for many. Monitoring the online presence of photos containing your image will become no different than monitoring your credit rating to protect your financial standing – only now you’ll also be trying to protect your reputation and character. And THAT is something to be concerned about.

Author: Ray Ganong, President, Applied Recognition Inc.

Posted in Face Recognition, Privacy, Sharing

Leave a Reply

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

*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>