What is facial recognition & how does it work?
Remember the way iPhone X ads caught your eye – Your face as your password to protecting the data and information on your phones was ultimate. They not only created a buzz with face recognition, but also literally handed over the might of face recognition or facial recognition technology to millions of the phone users to benefit and enjoy.
What exactly is Facial Recognition?
The ability to map facial features mathematically and store data as face prints to uniquely identify and verify a face based on facial contours is Facial Recognition. It’s a biometric based software application technology that functions based on nodal points of a human face. The values associated with these nodal points help to identify or verify each unique face. One can accurately and quickly find faces against captured data; however, technology is evolving with new 3D modelling approaches to better assist in overcoming current constraints
The Facial Recognition Advantage
Facial recognition is non-contact technique not requiring any interaction with user/person. This adds an edge for surveillance, security and monitoring to map time and presence that can be done from a distance at a comparatively low cost.
One needs to remember that the technology is light-sensitive and less favourable in low light and less effective when facial expressions vary. To understand these constraints, one needs to comprehend how facial recognition works.
How does it work?
Every person’s face is unique like a fingerprint, and the technology mapping this uniqueness behind facial recognition is complex, but the process is simple. It’s akin to your process of identifying a face – you get familiar with the facial features like eyes, nose, mouth and they come together to help you recognise. It’s the same here but on a grand algorithmic scale.
Your picture is captured from a photo/video or live; the facial recognition software reads the geometry of your face – like distance between your eyes, forehead to chin and identifies facial landmarks (software can identify 68 or more landmarks) and facial signatures are created.
The facial signature now a mathematical formula is then compared to other facial image for search and finally a determination is made.
This basic facial recognition process looks for faces to identify like the instagram filters – they seek features like eyes, nose, mouth and then use algorithms to lock a face and determine the direction it’s looking at.
The Face ID on your iPhone measures the distance between your facial features and every time you look at your phone to unlock, the camera looks at you – measures and confirms your identity.
The process to identify a stranger is similar but on larger scale – algorithms of that face is used to compare and confirm across a huge database that runs into millions. Thus, facial identification in this situation is difficult and hence different technologies are used to identify a face.
Omnipresence of 2D Imaging Technology
The most popular technology that facial recognition relies is 2D images for its sheer convenience. Mainly, a whopping majority of photos taken and available in databases and open source intelligence (like Facebook etc) are 2D images.
However, 2D images have 2 strong constraints.
1. They are not precise as these flat images lack identifying features like depth – the length of your nose, etc.
2. They are not clearly visible in dark or shadowy lighting conditions.
3D images with IR Camera for depth and clarity
This is achieved with a technique called lidar – The 3D camera blasts a harmless IR matrix – simply puts a wall of laser; that reflects from your face. The IR camera picks this reflection and measures how long it takes for each bit of light to return. Obviously, the ray from your nose returns earlier to say your eyes or ear. It then creates a unique depth map of your face. This 3D image when used alongside the 2D photos significantly increases the accuracy of facial recognition software.
Thus, Lidar imaging is this 3D indentation your face has left in the IR Mesh; imagine – like a mask of laser – where your nose is deeper than your eyes.
Better Facial Recognition with Thermal Imaging
The second constraint of low light, low visibility is well addressed by using a thermal imaging camera. Thermal imaging cameras rely on IR Light that the objects emit. Warm objects emit IR Light, more advanced and expensive technology is available to detect subtle temperature difference too.
Some of the simple ways used to identify a face with thermal image
Use of multiple photos
Thermal imaging camera captures more than one photo of a subject. Each time it focuses on different spectrum of IR light like long, medium or short wave to better gauge details.
Face print with blood vessels map
The blood vessel map is unique and can help formulate facial fingerprints. One distance finds out that distance between organs or identify scars and bruises too.
This is done by creating a dataset or a composite image. It is created by using many IR images. The final image that is formed can be used to compare and search a face from database.
This technology doesn’t work well in daytime thus it is still mostly used by military. However, as is case with new potential and evolving technologies, facial recognition has its own setbacks, but researchers and manufacturers are working to enhance the usability and accuracy of the systems across industries.