Facial recognition is future technology incarnate—all those years of watching science fiction, watching a character have their face (or iris) scanned is hardly fiction anymore. Today, facial recognition technology can be deployed in a variety of ways and for many purposes, from security to social media.

Just like any other emerging tech, facial recognition is a double-edged sword in its use. In some cases, there is concern about the powerful tool that facial recognition gives to authorities to mass surveil populations. In other situations, however, it’s helping victims or being used in completely novel ways for academic research.

Here's a look at some of the ways facial recognition technology is currently being used and the methods used to help the technology succeed.

 

A Brief Introduction to Facial Recognition Methodologies

Typically facial recognition is deployed using either 2D or 3D methods. 2D systems use either a geometric or photometric process.

Geometric 2D facial recognition identifies key facial features and determines the distances between them to identify a unique face. A photometric approach analyzes an entire photo and compares it against a database to make a statistical determination of identity.

Geometric methods can fail depending on how a photo is taken. By comparison, photometric methods utilize an already existing database so an analysis of a person’s face must exist. 

In 3D methods, key facial features are identified and measured by distances relative to each other, but also by the depths of those features. This method is even more precise than 2D facial recognition methods and works in a variety of lighting conditions and angles.

 

An example of a 3D face model used in facial recognition systems. Image courtesy of Advanced Source Code.

 

There are also completely novel approaches such as skin texture analysis, which can distinguish the difference between identical twins (something even 3D facial recognition can fail at). This is achieved by analyzing the skin texture of an individual and creating a unique mathematical signature based on pores, facial lines, and marks.

 

TrackChild Aids in Rescuing Missing Children in India

It is estimated that nearly 500,000 children are reported missing in India every year. India is a country with a population of over 1 billion, so being able to review each case and meaningfully identify each child can be an overwhelming task by traditional police methods. Backlogs of missing child reports can put children in danger of being neglected, abused, or exploited.

 

It is estimated nearly 500,000 children are reported missing in India every year. Image courtesy of NDTV.

 

As part of an experimental project, the Ministry of Women and Child Development partnered with the National Informatics Centre to create the TrackChild database. This database is first given photos of children reported missing and, when deployed, it can be used in conjunction with facial recognition technology to identify missing children. The database can also be shared and accessed across departments in India.

In a test run, TrackChild was provided 45,000 photos and was deployed in Delhi. As a result, nearly 3,000 missing children were positively identified in a matter of days—a feat unlikely achieved without the system. After positively identifying a missing child, steps are taken to reunite them with their family or guardians. The ministry plans to use the system to help map out locations where children are reported missing the most.

The exact method being used in TrackChild for facial recognition is not known—but, based on available information, it likely employs a 2D geometric analysis.

 

China’s Sharp Eyes Program Identifies Criminals in the Crowd

China has made its Xue Liang—or “Sharp Eyes”—program no secret. The name is derived from a slogan of Mao Zedong, who chaired the Communist party in the 1950s and founded the People’s Republic of China, to keep sharp eyes on other citizens. 

 

Example of variation between and within classes. Facial recognition technology scans crowds in public and private spaces. Image courtesy of the Department of Homeland Security.

 

Sharp Eyes is a database that combines the use of public and private security cameras to continuously monitor its citizens, identifying each face and storing it in a national database. Police in some Chinese cities now wear smart glasses that automatically scan faces in crowds and alert an officer when someone of interest is identified.

High resolution (ultra-HD and 4K, night vision-enabled) street cameras are also being used to scan drivers and identify anyone who currently owes fines or is not licensed to drive. This includes individuals who had their licenses revoked for drinking and driving. 

While the system is being promoted as something that is maintaining security and order, it poses challenging questions about the concepts of privacy and the limits of governmental control.

These issues are not unique to China, however. In the US, the FBI uses facial recognition technology in its NGI (Next Generation Identification) System. It's also possible that police bodycams may now also enable facial recognition in an attempt to bring similar functionality to US law enforcement.

 

Understanding the Universe Through Image Recognition

What does facial recognition have to do with identifying galaxies? More than you might think.

Convolutional neural networks (CNNs) are used by many facial recognition systems, including Google’s own CNN. Now, researchers have successfully used CNNs to identify images of galaxies and their phases—in particular, the “Blue Nugget” phase (including pre- and post-). This phase is an early stage of galaxy formation, when gases are contained in a relatively hot and dense region that gives off blue wavelength light.

The approach is similar for both facial recognition and galaxy identification in the sense that the system is trained to identify certain features of an object to identify it against a database. 

The galaxy identification system was trained using images from VELA simulations of the phases of interest. Then, actual images in three different filters from the Hubble Space Telescope’s Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey project (CANDELS) were used and classified. The pixel distributions of the images were analyzed, and the most important information is then automatically identified by the system.

 

Examples of images from the CANDELS project. Image courtesy of the Galaxy Catalog.

 

With the limitations of the simulations, the team was surprised with the results of the experiment. The system was able to classify Blue Nugget phases from CANDELS images with roughly 80% accuracy.

 


 

What other interesting or exciting applications of facial recognition technology do you know about? Let us know in the comments below.

 

Comments

0 Comments