- Identity Illustrated
- Posts
- Issue 5— Understanding the Relationship Between Context and Identity Authentication
Issue 5— Understanding the Relationship Between Context and Identity Authentication
In this issue, we will discuss the role of context in identity authentication and how businesses can improve results when scaling.
Over the last two decades, facial identification systems have become incredibly efficient at authenticating people’s biometric data, i.e., proving that a person is who they say they are. Smile ID’s SmartSelfie, for instance, can match faces with a 99.8% accuracy regardless of skin color.
A combination of technological advancements, improvements in algorithms, and the availability of high-quality datasets for training has catalyzed this increase in the efficiency of facial identification systems.
In an ideal world, facial identification programs would demand high-quality image or video data from users to verify their identities. But the world is not ideal. Much less is Africa, where varying levels of digital literacy, bandwidth access, and device specification make it nearly impossible for businesses to maintain control over the quality of images that users provide.
Although there are so many imaging factors that affect an algorithm’s ability to verify identity, perhaps the most important one, and one that businesses have control over, is context.
In this issue, we will discuss the role of context in identity authentication and how businesses can improve results when scaling.
Enjoying the Newsletter? Invite your friends to read it too
Context: More Than Just Biometrics
Modern-day identity authentication solutions rely on computer vision algorithms, which allow computers to interpret and make decisions based on visual data like images or videos. These algorithms extract unique identifying features from the biometric data presented by the user and compare them with a template to see if they match. In the case of image and video data, this technology is typically capable of proving that the person in the media data submitted is the same as the person on a document.
However, fraudsters are an ingenious bunch and have quickly taken to spoofing to trick biometric solutions into verifying fake identities. During these spoofing attacks, fraudsters attempt to pass themselves off as someone else. The exact mechanics of the attack can vary from simplistic attacks using printed-out images of another person to AI-generated videos mimicking the mannerisms of a real (or sometimes fake) person. Regardless, the end goal is to fool a biometric solution into thinking the image or video presented is legitimate.
To address the challenge of spoofing, biometric authentication solutions have evolved to account for more than just biometric data when interpreting an image or video. One such consideration is the context in which the media data is captured.
Context in biometric authentication refers to the inclusion of a substantial region around the face or document in the image or video.
Context in biometric authentication refers to the inclusion of a substantial region around the face or document in the image or video. This region provides important information that helps ID authentication algorithms (document or biometric) make better-informed decisions on the authenticity of the data.
Identity authentication algorithms go beyond just capturing a subject. They take into account the spatial and situational environment of the subject. Shadows, lighting conditions, and background elements play a crucial role in validating the authenticity of the document or the biometric data.
In both document and facial verification, the ability to authenticate increases with more visual context. An image of a person on a phone with a high-quality camera will capture all their biometric features. One way to distinguish between an image on a phone and an actual live image is by capturing the context around the picture.
The need for contextual image capturing is even greater with document verification, as many authentication cues live at the borders of identity documents. The trimming of an identity document, for instance, can be used to infer whether it has been tampered with or not. Watermarks, guilloche patterns, micro printing, and other features are routinely included at the borders of identity documents to make it easy to detect tampering.
Context clues can live at the corners of ID documents
Improving identity verification with context
As identity verification algorithms rely heavily on context to authenticate data, companies may find that pass rates drop for identity verification jobs without proper context. This may especially be true of companies that rely on APIs instead of SDKs to perform verification tasks. Where available, SDKs like Smile ID’s already account for context in image/video capturing.
Low pass rates during ID verification inadvertently lead to a laborious verification process and significant churn — which is a business problem. By improving context during the data-capturing process, businesses can increase pass rates and reduce time-to-value for customers.
One way businesses can improve context during data-capturing media is by including precise and simple instructions. When building solutions, developer expectations and user behavior do not always align.
As such, developers need to provide easy-to-understand text instructions to users when building data-capturing pages. This text can also be accompanied by instructional images or videos demonstrating how users should take a live image or video.
Depending on the ID verification provider, businesses may also be able to provide extra contextual information like the metadata of the image/video. Where possible, this data should be captured and shared with ID verification providers to enable accurate authentication.
Finally, when sending data over APIs, developers may consider resizing or compressing the data so as to achieve higher-speed turnarounds. However, machine learning models like the ones used in image verification can be sensitive to the smallest changes in images even when imperceptible to the human eye. To avoid images getting flagged, minimal resizing or compression should be done on images before they are sent.
In The News:
Bank of Uganda says it can’t protect citizens from AI fraud, calls for individual vigilance — The Monitor
Editor’s Note: AI fraud is a fast-rising threat to financial institutions across the world. While the Ugandan regulator has acknowledged its current limitations, it has also pledged to issue new cybersecurity guidelines and provide citizens with information on how to protect themselves.
South African mobile networks urged to increase defense against SIM swap fraud — Africa.com
Editor’s Note: Although SIM swap fraud dropped by 11% between 2021 and 2022, there are still thousands of cases every year. The article calls for more stringent measures, including biometric SIM registration, for better security.
ABSA Bank introduces new wallet for WhatsApp banking in South Africa — CIO Africa
Editor’s Note: The new wallet, introduced by South Africa’s leading bank, looks to improve financial inclusion by leveraging WhatsApp. Users will be required to take a biometric registration and create a PIN to access the wallet at each use according to ABSA.