Facial Recognition – Present and Future [Tech Tuesday]
“How are you being watched?” This was the opening question to a session last week in Southern California on facial recognition. Taking place at the IT Expo last week in Anaheim, the session “Facial Recognition – Changing The Communications World”, dealt with how facial recognition has been used and how it might be used. “Big brother has gotten a lot more involved in what we’re doing and how we’re doing it,” observed Michael Dundon of Mera Software Services, who delivered the talk.
Dundon pointed out that the current largest markets are social media tagging, passport control, criminal investigation and fraudulent applications (and they used facial recognition even as early as Super Bowl XXXV(!)). However, where is facial recognition going? Dundon pointed out some potential future markets could be school campus for security, Open Sesame, access control, authentication, high end retail, and banks and credit cards. Indeed, Windows 8 already has it for logging in, Windows 10 also has it and improved upon it.
A couple of questions that Dundon proposed and answered were: “How does it work?” and “What is the actual purpose of it?” Apparently, there are several different ways of getting pictures at the outset, in order to match up video or picture-capture faces. Mugshots are the most sought-after version, usually jpeg compressed, since they are clear. However, web cameras make up 14%, which are usually people who are actually interested in sharing it, but these pictures are very inefficient and not compliant with a lot of standards, coming in at about 7kb per picture. Once one gets those pictures, one takes those pictures and populates the database. Then, one uses ISO/IEC 19794-5 and then there are a variety of algorithms to be used for facial recognition, with three of them being the more prominent amongst them.
Once one figures out where one wants to capture the pictures, then one needs to deploy them – this could be a doorway, a sales counter, whatever. Then, once it’s captured, one needs to work on the vectors, by removing all of the soft features (hair, tattoos, etc.), then just look at the actual face, such as chin, nose, etc. (although glasses could distort where the eyes are), then, an analysis is started. Once the analysis is started, there are several factors that need to be taken into account. The final stage after the analysis is identifying if the person is in the database or not. If they are, then it’s identifying who it is (if they are not in the database, they can’t be identified).
With respect to accuracy in pictures, it is higher in adults, since adults don’t change as much within 10-15 years. However, children may last a year or 6-8 months, depending upon when a picture is taken, then it needs to be updated. Also, some software will only do well with someone’s face straight-on, while some will allow for up to 30% movement, whether turning to the side or up or down. Also, as to what the capture rate is for these different softwares, Dundon said, “it all depends upon the algorithm and the methodology that one uses.”
A non-commercial use that Dundon pointed out at the end for facial recognition was in the aftermath of the terrible floods in Houston this summer. There were lots of pets lost in the flooding, while many of them ended up in animal shelters. So, what was done was to have the pet owners submit pictures of their pets to a database and matched them up with pictures of animals in the animal shelters, in order to reconnect them.
How facial recognition will continue to be used in this day and age is still changing and should remain a dynamic area.