Facial Recognition

Reviewed by: Navaneeth Kamballur Kottayil
Last Updated: August 14, 2020

Definition - What does Facial Recognition mean?

Facial recognition is a new technology that's being built into all sorts of applications, from airport surveillance kiosks to social media engines.

It's also one of the more controversial technologies being pioneered today, as it sets up deep questions regarding security versus privacy rights, and how these facial recognition applications can be safely and fairly applied.

Facial recognition is also known as face recognition.

Techopedia explains Facial Recognition

Modern facial recognition is clearly dependent on specific technologies and algorithms that we've built during the machine learning and artificial intelligence era of the early 21st century.

Specifically, most cutting-edge facial recognition programs feature a type of neural network called a convolutional neural network (CNN). The system uses convolutions as well as other algorithm work in successive stages to do complex analysis of an image, and even identify people, animals, objects or settings through advanced analysis.

How does a convolutional neural network do its work?

One primary piece of functionality in the CNN is feature detection. Using elaborate algorithms, the program will break down an image through color shift and local analysis of group pixels to find features — for instance, in the human face, features like noses, ears, eyes, etc.

The same facial recognition neural networks will often utilize ratios — such as the ratio from eyes to hairline, from ears to nose, or other stock facial ratios that can help with facial recognition. The ML program can use the uniqueness of each face to learn how to identify the individual using existing data and extrapolation principles.

Experts characterize the layers of a CNN as doing “Sherlock Holmes detective work” on an image. Other aspects of this investigative work involve max pooling, where the machine learning program just keeps the most relevant information while discarding useless data, and non-linear pattern recognition algorithm such as ReLu, which is sometimes described as an "activation function" within the network.

Consecutive rounds of convolution/ReLu/pooling create the combined effect. Techniques like striding and padding allow the program to “scan” the topography of an image to perfect its methods.

So, with all this winning technology the facial recognition engine can be surprisingly adept at learning how to recognize a particular individual’s face in a crowd.

What are the concerns?

Primarily, companies that have used abundant public Internet images to pull together training sets for sophisticated facial recognition programs face blowback and resistance from some of their customers, including law enforcement departments, and from U.S. legislators, consumer advocates and citizens at large.

In other words, people aren't always comfortable with allowing these technologies to work on identifying them from digital photographs. Context matters, and so does applicable privacy law. The idea of “ethical AI,” promoted by top innovators and industry experts, applies to facial recognition in a particular way.

The above shows a bit about how facial recognition works technically, and how it's being applied in our societies. Both of these will be significant in future ML/AI development moving forward, as facial recognition functionality gets increasingly built into items like smart doorbells and other devices potentially useful in broader surveillance.

Share this: