Back to top

A Digital Book of Faces

The ability to see with facial recognition

Have you ever thought you saw someone you knew in a crowd only to realize it was actually someone else? Even if you’re one of those people who is “good with faces”, you probably can’t remember and distinguish between millions of them. But a computer, or an algorithm behind software, has been doing this for years now.

In perhaps the most famous case of everyday image recognition, Facebook has been identifying your friends and family in any photo that gets uploaded. Considering how long they’ve been perfecting this technology and the amount of data their AI behind facial recognition has had to learn from, it’s no wonder they hold the current accuracy record of 97% . They’re getting better all the time and they’re not the only ones.

The development of image recognition techniques has created opportunities almost everywhere. The implications of what can be accomplished when a computer can “see” as a human does has attracted large amounts of research attention over the years, making image recognition one of the most popular subjects in the field of AI. Image recognition can utilize many aspects, including multilayer perceptron (MLP), convolutional neural network (CNN), recursive neural network (RNN) or long short-term memory (LSTM) neural networks. In fact, a hybridized approach is typically the most common path taken.

Law enforcement and security agencies are now using facial recognition in the field to track and identify suspects. Facial recognition can unlock your phone or be used to make purchases. Hotels use it to welcome guests and some schools are even taking attendance this way. Although many algorithms are continuing to be explored and implemented, deep learning techniques utilizing neural networks have received the most attention.

An AI can be used to read unstructured data from images or be used to improve graphical user interface (GUI) interaction.

For example, we know that image recognition will play a critical role in training an AI by “watching” a user interact with an application. An AI can create heat maps of user engagement or translate one workflow to another in the event of a web redesign. An AI can translate images in real-time or make command decisions in near real-time.

Image recognition can also help validate signatures on forms, clean images of noise or corruption, and even restore images that have been damaged. In short, if it is something we use our eyes to understand, an AI can learn to do it too.

 

References

Forbes

Blog Author

By Brad Mascho
Chief Artificial Intelligence Officer, NCI