Although face recognition technologies have been in use successfully for several years, the debate regarding benefits and potential drawbacks continues. Public concerns around privacy and government adoption have increased as the technology has become more widespread. Much has been reported about this.
Face recognition technology vendors are adjusting and responding to the issues raised, as are the customers implementing these tools. There is a sweet spot where we can derive the benefits that this technology brings, while minimizing the potential for misapplications. Technology advancements and increased deployment flexibility are helping reach this desired sweet spot.
The benefits of face recognition technology
The list of beneficial use cases is long, and the benefits that the technology provides are difficult to discount.
In the security space, face recognition technology is indispensable. The cost of alternatives (particularly the application of human capital) is prohibitive. Without face recognition systems, security would be more expensive and probably achieve less.
There are also many other application areas. At its core, the technology provides a new way to automatically interact with devices and data, and this can make life safer and more convenient. Unlocking your phone with your face is but one example in the convenience category. Many others in the very helpful category include applications such as identifying missing children and lost elderly, verifying a patient’s identity in order to avoid medication mismatches, verifying identity at the ATM machine, etc. The list is as long as we let our imagination run.
Avoiding the potential for misuse
There are many technologies in use today that could be thought of as potentially promoting the advent of an Orwellian world. The reality is that we are being tracked regularly and voluntarily even without the use of face recognition tools. Our phone’s GPS is tracking our every move on the street. Online, our clicks are being tracked and correlated. At home, our IoT devices are also keeping tabs on us and what we do. We are doing this voluntarily. In context, misuse of face recognition tools would be at best an incremental contributor to this picture, and probably far from the most significant. Privacy is a very important matter, and issues surrounding it should be viewed holistically and comprehensively if we are to succeed.
On-premises face recognition software: minimizing risks, maximizing rewards
In today’s marketplace we find that some face recognition tools are offered as a service on the web, and some are available for on-site deployment by the customer directly. There are advantages and disadvantages to either approach, but the on-premises approach avoids important security/control issues that are particularly significant in government implementations.
Web offerings consist essentially of an API (Application Programming Interface) that is used by a programmer to submit a face image and obtain a response (for example, data that characterizes that face).
These systems tend to be opaque to the end user by their very nature. The images are processed on servers that are owned and controlled by the service provider, creating gaps in the chain of custody for the data. One cannot be sure what happened to the uploaded image, leading to a series of questions: Where was it processed? Was it cached or altered? Did anyone else see it? What happened to the results returned? Was the data repurposed? The list of questions is lengthy.
On-premises solutions can resolve these vulnerabilities and serve as a great alternative to purely web-based solutions. The fewer “hands” on the images/videos/data, the lower the possibility of leakage and misuse. In our view, law enforcement and government organizations that deal primarily with highly sensitive data should be in full and sole control of their data and processes.
Recently, the Amazon Board of Directors rejected two shareholder propositions to stop selling face recognition services (Rekognition) to police and government organizations. The episode highlights an important issue for long-term customers. Metadata that was generated with a service vendor’s software can typically only be used with that software. If the vendor withdraws from the market, the customer’s multi-year investment in the data previously generated goes out the window. This is a pitfall that can be avoided by installing an on-premises face recognition software. The customer remains in control.
Historically, face recognition technologies have evolved from the perspective of trying to match a face in conditions where the person wants to be recognized (for example, a door access control point). Scaling this approach to conditions “in the wild” or where the subject is an adversary, however, does not work nearly as well. In the past few years, face recognition technology has improved significantly and can now achieve successful recognition under a broad range of challenging conditions. For instance, contextual information in the image can help to refine the scope of what a user is searching for. This is a big leap in computer vision leading us closer to automated image understanding. The desire is to not just find the proverbial “needle in a haystack,” but also do so when the “haystack” comprises just good needles and one is interested in a very specific needle. It’s important to be able to look at the context of the image to pull up what the user needs. The software must look at everything in the image: objects, text strings, scenes, etc. A face in an image may be partially obstructed, but other aspects (clothing, etc.) can contribute to the confirmation of that match.
In situations when there is an abundance of good face matches, only few of which are relevant to the investigator, this is another area where context from the image can help filter results (for example, a store name in the background where the text was also automatically detected and recognized, a landmark building in the image, a type of flower in the background, etc.).
As the technology continues to evolve, the new capabilities can help make some of the concerns for misuse and mis-identification go away. We are also getting smarter about how to implement it effectively because we want a better world, not an Orwellian one.
The views expressed here are the writer’s and are not necessarily endorsed by Homeland Security Today, which welcomes a broad range of viewpoints in support of securing our homeland. To submit a piece for consideration, email HSTodayMag@gtscoalition.com. Our editorial guidelines can be found here.