Will Face Recognition Software Ever Match the Human Brain?

New biometrics technology is making identity theft more difficult everyday. Although face recognition software is amazingly accurate, it could still learn a few things from the human mind.

Although biometrics isn’t the end of identity theft, it is making huge strides in security. Modern face recognition software can identify someone even from a distance and match his or her face to a known person. This technology is quite advanced but still has a long way to go.

Current state-of-the-art face recognition software uses lines of the face and distance between points on the face to match identity. And it is very good at it. Face recognition software has a margin of error of less than 1 percent with a clear view of a persons face. This is equivalent to the human minds’ ability to recognize people it is only the most familiar with. And human recall capacity doesn’t even come close.

Although you may not think a person could compete with face recognition software in matching facial features, we can do some things that a computer still can’t. Engineers are trying to put these abilities into the next generation of face recognition software.

What can you do that face recognition software can’t? You can deal with changes or variations in appearance far better than face recognition software. This is because you can take into account a whole realm of circumstantial cues we haven’t been able to program into face recognition software.

When it comes to child aging, gaining or losing weight, or face contortions, face recognition software is almost always stumped. But you can look through someone’s photo album and know exactly who people are in a picture without much effort, even if they’ve changed significantly. How do we do this?

The answer is we use a lot of cues that aren’t even in the picture.

If you know you haven’t seen a young person in a very long time you might expect them to appear a little differently. Generally, you know the visual changes that take place when a person gets older – longer distance between eyes, smaller eyes in relation to the rest of the face, more prominent bone structure, etc. Face recognition software can make these predictions, but there’s no formula for how everyone ages. The more time that goes by and the younger a person was when a picture was taken, the harder it gets for people and computers to accurately recognize them now. And what if you don’t know when a picture was taken?

But you might have some other tricks up your sleeve that come from living and drawing from a huge store of random frames and ideas that we may never be able to mimic in face recognition software.

If you know what the person’s parents or siblings look like, you may expect him or her to age similarly. This can greatly help recognition. You might get clues about the time setting of a picture you’re looking at, such as a 1980s hairdo or a pair of He-Man underwear. This could lead you to assume an approximate aging effect on the person’s appearance now.

It may be hard to recognize someone in an old picture who has since lost or gained a significant amount of weight. But if we know others in the picture, it might bring to our minds a set of people connected with them who we might also expect to see in the picture. Then the difference in appearance may not be as difficult to traverse.

We make judgments like this unconsciously everyday. But these are things face recognition software may never be able to do on its own.

Our ability to judge changes in color is another advantage we have over face recognition software. Computers are good with light and dark but not with color. We can see hair go from red to gray and assume a change in age. If you see a picture of a girl you know, but with darker skin, you would probably judge that she had a tan. This may give face recognition software a hard time.

Another hurdle that face recognition software hasn’t quite mastered is the appearance of features not normally part of someone’s face. These might include scars, jewelry, glasses, or sucker sticks protruding from the mouth. These are common items we might not even notice on a person because we know what they are and we ignore them, but they can really foul up face recognition software.

These human abilities give us a much higher tolerance for variation in appearance. This is something engineers are working on to improve face recognition software. And humans are the model. For a human mind to recognize people and situations as we do everyday requires hundreds of unconscious split second judgments that rely on a lifetime of experience. Someday biometrics technology might be able to make the same circumstantial judgments we do, but not without a lot of work.

About the Author: Mat Moniker is a writer for Innuity.com and a biometrics expert. To learn more about the face recognition software on the market go to Fulcrum Biometric’s Website.