Taking Great Selfies in Africa
As Quartz and Forbes reported back in June, Africa’s top handset maker Transsion won the hearts of its customers was by listening carefully to them and building products they wanted.
This year, the latest iteration of this customer-focus took the form of optimizing the cameras on their new line of Tecno, Infinex and Itel smartphones to work especially well for faces of color — capturing rich, vibrant selfies under a wide variety of lighting conditions and for a range of darker skin tones.
At Smile Identity, we’ve also realized that the quality of our facial recognition depends directly on the quality of the images our users take.
To deliver for our Partners and users, we have to make face recognition work no matter what environment they find themselves in, whether they are ordering a taxi at night in Lagos or applying for a loan from the back of a bumpy Matatu in Nairobi’s morning rush hour.
Having worked in identity and financial services for years, our team knows that the best KYC process is one users don’t have to repeat. So in June we analyzed the results of over 1 million images captured using our SDK and saw a substantial number were being rejected for various image quality reasons. The most common culprit was poor lighting, but blurriness was also a factor, especially on low-end phones with slow frame rates.
In order to prevent users from sending selfies that wouldn’t be useable for opening bank accounts or authenticating a transaction, we added on-device blur and brightness checks to our SDK, with real-time feedback to users so they could adjust before sending a selfie to our Partners.
These real-time prompts tell users if there is not enough light, or if they need to hold the phone still to avoid blur.
But just like Transsion, we went a step further.
While training on our 1 million images, the majority of which were faces of color, we were able to develop algorithms that optimize the use of a camera’s “Exposure Control” for face recognition. This allows us to finely control the sensor on a camera phone and detect more light (more photons) when needed while minimizing image noise, allowing us to get greater facial detail from an image being captured.
This Smile ID Exposure Control takes advantage of “Metering,” a function built-in to almost all new smartphones and roughly half the installed base of smartphones we’ve seen in Africa. Metering works by adjusting brightness settings for a defined area. Anytime you tap on a camera capture screen with metering, you focus the camera, creating a defined square for the sensor to optimize.
Our technology works the same way but without a user having to do anything. Smile ID Exposure Control works in conjunction with our blur and brightness checks. For devices with metering, instead of just telling the user they need to move their face closer or find more light, we can go a step further and actually adjust the sensor automatically based on a face’s color and the lighting conditions around it. The resulting images are optimized for ambient lighting and skin tone resulting in clearer images and better face recognition.
Smile ID Exposure Control is part of our latest mobile SDK (SmartSelfie™ V6) and extends the kind of optimization for faces of color that new Transsion phones have, to all smartphones with image metering.
As a developer you don’t have to do anything. Our SDK checks to see if metering is available on a device and adjusts the exposure control until we can detect enough details in that face to know that our face recognition will work. This is critical to making face recognition work on low end phones or in challenging conditions. And if the image is too bright (for example, a pale white face on a sunny day), we can also decrease the exposure programmatically in the opposite direction, until we properly detect a face.
To try our new SmartSelfie V6 SDK, sign up for a Smile Identity developer account.
Built with purpose in Oakland, Cape Town, Joburg, Lagos, Nairobi and Kigali — the Smile Identity Team
