Can You See Me Now?

Once again, it’s graduation season with the parties, the ceremonies and the yearbooks (are yearbooks still available in hard copy?). I recall elementary school class pictures where we lined up in rows sorted by height. In high school the lineup was less orderly, and many skipped class pictures at college graduation because we had places to go, things to do, careers to begin, and families to raise.

On occasion, I look at those old pictures and recognize most of the faces, yet I wonder what they did with their lives and how they might have aged. Shortly after graduation, I would have recognized and greeted them by name, but how many classmates could I recognize today if I passed them on the street? Both aging and growing divergence of our shared experience would limit how many I would recognize.

Our natural ability to associate facial images with names is limited by time. Today, facial recognition by technical means has been improving, and in some instances, it is near perfect, especially in closed systems that recognize the carefully posed-faces of a limited number of “cleared” employees.

In less well-honed databases, people may not be carefully posing, and their pictures may be pulled from different databases. Images may be grainy, or there may be just a single image per person. The pictures may be taken from different angles, or there may be a sample shortage of images with different racial and ethnic traits, which could increase the degrease of precision. A database with those faults can result in imperfect training of the recognition system. Inaccurate recognition can result in biases, and the public’s rejection of a facial recognition system.

Better images in databases and more extensive training for the system can improve recognition accuracy. However some privacy activists object to facial recognition due to fears of the nefarious uses that government, politicians and law enforcement might employ facial recognition to achieve. Some individuals have deployed facial recognition to protect themselves from known troublemakers, yet others object to such personal protection measures. Many objections are rooted in paranoia, but there have indeed been some rogue examples of facial recognition such as in today’s China and in dystopian tales like George Orwell’s 1984.

Today, there are large networks of public safety camera networks in the UK and in American cities such as Chicago that disturb some people. The presence of street-level cameras that police use to monitor public safety should not be considered a genuine threat. We do not live in Beijing.

We have seen this syndrome before. Some would like to ban the tools that both criminals and innocent people use – knives, rifles, automobiles, computers and so on. Instead, our focus should be directed to curbing criminal behavior and criminals, not toward limiting everyone’s use of benign tools.

For some, a database of facial images is considered a violation of privacy by some of those in the database. The stored images may have been be collected with permission, or they might have been collected as people pass by on a public street, or people may have taken part in a group picture – for example, in a graduation photo. Unless people take active steps to hide their faces while in public (e.g. sunglasses, fake beards, veils), they are de facto allowing passers-by to see, remember, or even photograph their face. Image database operators need other information such as a name to make the picture, and ultimately the database, most useful.

There are complications that may create shortcomings in image databases, such as laws that prohibit photography of public gatherings, or laws that force removal of some individual’s data. One law that limits the use of personal data that may have been gathered lawfully is the “right to be forgotten.” That law applies in the EU to collected online material. It empowers an individual to force a database operator to remove some or all of the individual’s data from public retrieval. In effect the right to be forgotten becomes a tool to choose erasure of a sordid past. It is unclear whether the right to be forgotten will migrate from Europe into the U.S. and whether it will then be applied to image databases.

The U.S. laws protecting privacy are incomplete from public, commercial, and law enforcement perspectives. Neither the privacy protections that many want nor the public safety exceptions that protect us from criminals are available.