PETALING JAYA – With the global push towards digitalisation, it is now more important than ever to equip ourselves with the necessary knowledge on high-tech solutions — including facial recognition e-KYC services, which have become more prevalent in our everyday lives.Through WISE AI’s regular publications, we have explored the different types of facial recognition methods, as well as their strengths and use cases. However, the question inevitably leads to this — How does WISE AI conduct its own facial recognition?
e-KYC and its needs
WISE AI primarily uses its facial recognition systems to facilitate e-KYC services, namely customer onboarding for the telecommunications and BFSI industry (Banking, financial services and insurance).
Due to its functionality, a few key parameters have to be set beforehand:
Keeping the above parameters in mind, here are the steps involved in WISE AI’s facial recognition operations:
Step 1: Face Landmark Detection
While some consumer devices, such as the Apple iPhone, have specialised camera systems to capture 3D facial data, a vast majority of devices capture facial data through 2D images — which is where face landmark detection comes into play.
It involves using computer vision applications to detect key landmarks on the face and tracking them — landmarks such as mouth, eyebrows, eyes, nose and jaw.
While it is possible to use a pre-trained model for face detection, WISE AI employs its own deep learning-based algorithms, to ensure better detection accuracy for Asian facial features, as opposed to the models that are trained using Western facial data.
Step 2: Face Liveness
Through face landmark detection alone, it is still possible for fraudsters to trick the facial recognition system by holding up a photo of another person at the camera. This enables unauthorised users to bypass the facial recognition system entirely.
Liveness detection algorithms attempt to solve this issue by detecting fake or non-real faces through various methods (which will explore in detail in the near future) — such as texture analysis, frequency analysis and variable focusing analysis.
Step 3: Pose estimation
Now that the user’s facial image data is properly detected and analysed as being genuine, the system then manipulates the photo’s position, size and rotation to fit a proper alignment that is standardised for the e-KYC process.
This ensures that all user-submitted facial data is consistent and usable — which is essential for archiving purposes. It is also an essential step in preparing the data to be fed into machine learning algorithms to train better facial recognition models.
Step 4: FAR and FRR
After going through this arduous detection and analysis process, WISE AI then uses False Acceptance Rates (FAR) and False Rejection Rates (FRR) as metrics to evaluate the effectiveness of a facial recognition model.
These two terms describe the rate of unauthorised persons that are incorrectly accepted and authorised persons incorrectly rejected. The key is to strike a delicate balance between these two rates — as one represents the level of security and the other represents user convenience. Reducing the FAR is likely to increase FRR, and vice versa.
Regardless, WISE AI holds the stance that the FAR and FRR should not be changed invisibly and must be made transparent to clients — a notion that many security experts agree on. For e-KYC purposes, where security is paramount, altering both FAR and FRR can be detrimental to the client’s operations.
Facial Recognition: Beyond tech
Even though facial recognition systems and e-KYC are both highly technical subject matters, the solution boils down to a simple goal — to verify a user’s identity. Hence, conversations around facial recognition should extend beyond just the tech, to more fundamental questions such as:
WISE AI is looking forward to playing an active role in promoting these discussions. Subscribe to our newsletter to get updates and insights by visiting the link here.