Introduction To Face Detection

Image credit:

What is face detection?

Face detection is a version of object detection that, as the name implies, detects faces that are displayed in a visual medium, often having to pick them out from other parts of the image such as landscapes, other body parts, or other people. While face detection itself has few direct uses, it is the first step in many key applications used in fields such as biometrics, law enforcement, and even entertainment.

Facial detection vs facial recognition

While these terms are often used interchangeably, they are not the same thing. Facial detection, as described above, refers to the detection of faces in a visual medium. Facial recognition on the other hand is a further application of face detection, and goes much further than simply detecting when and if a face is present. Once a face is detected the biometric data of the individual’s face is mapped and saved as a face print. This face print is then compared to a database of other face prints to find potential matches.

What is face detection used for?

If you’ve ever used a camera app on your phone, you’ve most likely already seen face detection in use. If you’ve taken a picture and seen a colored square appear around the photo target’s head, that’s face detection. When applied in cameras, this software is most often used to auto-focus cameras and identify if people have entered the frame, helping photographers take clearer and cleaner shots. Furthermore, face detection can tell a computer what part of an image or video to blur to maintain that person’s privacy.

Advantages of face detection

  • Privacy
    Unlike facial recognition, face detection does not store any identifying data. Face detection will typically only identify the presence of a face, and at most will use facial analysis to store general, unspecific data about the face detected such as age or gender. With cameras already nearly everywhere in our day-to-day lives, an addition of facial recognition software could serve as a nail in the coffin to whatever privacy we still have. While this may help law enforcement to detain criminals, it can still be used to keep tabs on the average citizen. Imagine a database containing hits from facial recognition software taken from every camera you pass by in your day, for every person in the area. It could, if storage allowed, feasibly have an accurate accounting for your whereabouts at every point of every day for an extended period of time. This isn’t even an unheard-of concept, as some governments are already implementing similar practices with phone records and other collectible data.
  • Improved security and safety
    Even using only face detection, having an automated process to determine a person is moving towards a secure area could be a large boon to security workers. Adding in the benefits posed by facial recognition, law enforcement would have a much easier time identifying and tracking down criminals. The TSA is even piloting a biometrics program that makes check-ins at airports less time-consuming and physically invasive by utilizing facial recognition to cut down on time spent.
    This technology is not necessarily limited to crime prevention either, as New Delhi found over 3000 missing children over the course of 4 days using facial recognition systems. This is just a drop in the bucket compared to the number of missing persons in India alone, but further refinements to the software and the ability to search for the missing persons earlier could reduce the number of missing persons significantly.

Disadvantages of face detection

  • Data Storage
    Face detection and all of its applications use machine learning to identify a face, and the storage requirements for such a process aren’t insignificant and would likely be unavailable to the average person. When taken to the further applications that require databases to match against or store additional data, the requirements can escalate quickly.
  • Vulnerability
    This is a particular concern with facial recognition applications in fields involving security, such as biometric locks or law enforcement. While these algorithms tend to be more accurate than manual checks, false positives and false negatives aren’t uncommon. While a false negative might just mean you need to scan again, a false positive can allow unauthorized users into areas that should remain secure, whether that’s your phone or something bigger. In addition, facial recognition software is particularly vulnerable to larger changes in appearance that a manual check would otherwise catch.
  • Algorithmic Bias
    These algorithms in their current state are far from perfect, but with boasts of classification accuracy exceeding 90% in some cases, you might think otherwise. The real problem comes from the disproportionately low accuracy when identifying minorities. The “Gender Shades” project in 2018 found that in particular, subjects who are female, black, and in the 18–30-year-old age group consistently have the poorest identification accuracy, with averages reaching as low as 20%.

Student at Holberton School