Subscribe to DSC Newsletter

Major Factors Keeping Facial Recognition from Mass Adoption

Artificial Intelligence and Machine Learning are accelerating and refining various industries. One of the most rapidly developing and progressive domains is Facial Recognition (FR). Its implementation in many spheres, from public security to retail and healthcare, only proves its potential.

 

 

Despite FR’s broad dissemination, there are many precedents where FR still makes mistakes. Media reports are filled with stories of FR’s racial discrimination, for example. The reasons for such failures vary, yet companies already using the technology have hope for its improvement and future benefits.

 

A National Institute of Standards and Technology research has shown that since 2014 FR technology has been refined more than 20 times. Moreover, according to Allied Market Research, the value of facial recognition technology is likely to rise to $9.6 billion by 2022 with a CAGR of 21.3% between 2016–2022.

 

While the continued development of the technology seems almost a given with companies striving to integrate it into their current solutions, what are the challenges that it faces and what are the prospects of the tech?

 

Facial Recognition Challenges

The issues facing facial recognition can be divided into two main categories—technical and ethical ones. While the former is just a matter of time and effort for the technology to develop, the latter may require a more holistic approach, including the knowledge of laws, social science, and even human psychology.

 

Technical Challenges

To understand the obstacles FR may encounter, let's take a walkthrough into how the system works and look at the challenges at each step.

1. Picture input

The system obtains an image of a person by scanning, taking a photograph or a camera recording. The main factors that influence facial recognition at this step are picture quality and resolution, meaning it is vital to have the right equipment and capturing mechanism in place.

 2. Face detection

The program decides whether there is a face in the image or not. This is an easy task for a human but a great challenge for FR. The system has to define if it ‘sees’ facial features in the image: two eyes, one nose, one mouth etc. For this, FR uses standard templates which it matches with a face. If any object resembles a human face, the system may mistakenly define it as one. The internet is full of comical examples of face swaping with objects, which, while hilarious, highlight the gaps in the technology.

3. Feature extraction

The program analyzes a person’s features to compare with the images in the database. It's one of the most challenging steps as the system needs to take account of the conditions (lighting, pose and face angles, facial expressions, presence of hats, eyeglasses, beard, mustache, makeup and even pimples, occlusion, natural aging and image resolution) in an image.

 

To solve the aforementioned issues, various techniques can be applied—from simple feature matching to 3D face modeling. These approaches consist of complex algorithms and often require lots of computing power and in-depth engineering knowledge.

 

4. Feature matching

In FR technology, this is the most complicated part. It’s also where the system truly shows its value. To match the features of an image with the database, the system relies on machine-learning techniques. First, the system learns to extract and classify facial features, then it applies its learning results to the image being analyzed.

 

The system may make mistakes as it depends on the ML learning and matching algorithms, database quality, as well as how well the features were extracted in the previous step. However, with technological progress, it has become possible to create an almost endless number of image-based data. Recently, a new approach to face recognition has emerged. Drawing on deep learning and use of neural networks, this new process became known as deep face recognition.

5. Face recognition

A lot of businesses aim to use FR for practical purposes in real-life situations, and this creates challenges for FR. For example, if the system is required to recognize a face in the crowd, it’ll have to deal with unconstrained environment where the background (multiple faces, lighting, quality, resolution, and a person him/herself: gender, skin-color, pose, head rotation, occlusions and movement) will create obstacles for successful face recognition.

 

Ethical Challenges

FR technology is widely used in many spheres, and it goes far beyond the consumer use of Apple Face ID and Windows Hello. It becomes part of more serious issues such as privacy and surveillance, resulting in changes in legislation and human privacy culture as a whole.

 

As the world of CCTV developed, with a camera on almost every street corner, people have become more or less indifferent to its presence. However, FR again highlights the discussion on the issues of invasion of privacy that plagued the early years of CCTV. Not least among these concerns is that the system isn’t always 100% accurate and could, in fact, be considered a violation of human rights.

 

FR still requires accuracy enhancement. For example, in 2018 Amazon’s system decided that 28 members of Congress resemble criminals. While this may sound like a joke, it gets more serious when a company is claimed to be racially biased due to misidentification. FR appeared to fail more often in face identification of darker-skinned people. This issue may result in more dramatic consequences for an ordinary person of color. On this occasion, the Congressional Black Caucus expressed his concern in a letter to Amazon CEO Jeff Bezos.

 

These issues are already motivating governments to step in and examine legal changes that need to happen. In March 2019, US Senators introduced a law called the Commercial Facial Recognition Privacy Act, aiming to regulate the use of data for facial identification. Now, companies applying FR will have to declare to their customers how they use and share the data. The law’s been enacted to preserve citizens’ privacy and protect them against discrimination and racial bias, drawing on democratic freedoms. The law is similar to the ones enforced in the European Union but specified for FR.

 

Takeaways

 

Facial recognition technology is advancing at a rapid rate. In coming years, we are likely to see great improvements in its quality as well as new developments in the laws governing it. Here are the three main takeaways for the future of FR:

 

  • For businesses, facial recognition is a promising and fast-developing technology that is already implemented in many spheres such as retail, banking, and travel..
              
  • Facial recognition implementation requires a comprehensive approach, not only from the technical perspective, but also from the side of human relations. Systems need to be put in place to manage the privacy issues surrounding the technology and its governance.
               

Facial recognition is a sophisticated system and its accuracy doesn’t always meet the 100% ideal. However, the development of Big Data and neural networks unlock new possibilities for the technology’s development.

Views: 349

Comment

You need to be a member of Data Science Central to add comments!

Join Data Science Central

Comment by Steven c Philpott sr on July 10, 2019 at 2:26pm

Well done.   A good read for policy makers that are on the parking brake about smart city innovation.

Videos

  • Add Videos
  • View All

© 2019   Data Science Central ®   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service