DeepFace: Facial Recognition Software & Privacy

April 15, 2014

Facebook recently published a paper on a new technology that it is developing called DeepFace.  DeepFace is a system that determines whether two faces in different photographs are of the same person by building a three-dimensional model of the face and rotating it to match it with the face at different angles.  The technology has 97.35% accuracy regardless of lighting and the angle of the face in the photos, which is extremely close to the human level of accuracy, which is 97.53%.
If implemented, the new technology will improve Facebook’s tag suggestion feature by identifying the correct friend to tag more often.  The current tagging feature was highly controversial and led to a probe by the European Union, which resulted in the feature being temporarily suspended.

The technology has 97.35% accuracy regardless of lighting and the angle of the face in the photos, which is extremely close to the human level of accuracy, which is 97.53%.

DeepFace is currently in the research phase and its exact use is not known.  It is expected that more information will be unveiled about the technology in June at the Conference on Computer Vision and Pattern Recognition Conference in Ohio.
Despite not knowing DeepFace’s ultimate use, the private innovation can provide law enforcement agencies with greater ability to identify and track suspects.  One recent example of where DeepFace would have been an asset to law enforcement agencies is the Boston Marathon Bombing, which occurred a year ago today.  A case study on commercial face recognition software, at the time of the bombing, found that the available software’s matching accuracy was not accurate enough for full-blown use by law enforcement agencies.
The biggest concern with DeepFace, as is with a lot of facial recognition technology, is privacy.  In 2012, the Federal Trade Commission issued recommendations that users be informed of a private company’s use of facial recognition software and given the choice to opt out.  Facebook currently follows those recommendations and is likely to do the same if it integrates DeepFace with its current tagging system.
Despite privacy concerns about the use by Facebook, there is also a concern law enforcement will be able to leverage the technology.  Currently, neither regulations nor guidelines are in place for the use of this technology by law enforcement officials.  Jay Stanly, a Senior Policy Analyst for the ACLU, blogged that the ACLU does not have concerns about using the technology to help police identify a suspect against mug shots of convicts, but the organization does have an issue with law enforcement using a public database such as a DMV database to identify suspects.  Stanly suggests that for law enforcement to have access to the DMV database, or a larger public database, a compelling reason must exist and a system of checks and balances should be established.  Stanly does not believe the FTC’s guidelines address the real issue of the matter, which he identifies as “whether the de facto anonymity that most people have enjoyed in urban areas since the dawn of industrialism will come to an end through the use of face recognition technology and ever-more-pervasive video surveillance cameras to routinely record everyone’s comings and goings.”  Whether these fears come to roost, there is still time for laws and regulations to be adopted before DeepFace and similarly technology reaches the market.