I'd suggest that facial recognition is less than 90% accurate, even with a frontal image, in a controlled environment, with good consistent illumination and without blur, distortion and other camera effects and the effects of skin tone/race/colour*.
Is the system using the biometric data in the user's Passport, the US immigration point image capture or an image captured elsewhere e.g. check-in / baggage drop and issue of a boarding pass (separate queue for those that checked-in online and printed their boarding passes, only have hand luggage).
If it's the automated booths as used at LHR, a potential nightmare; one will be broken, someone will have trapped themselves in the second, and the old lady with the sun hat has confused the third.
If it's a person looking at an image panel and comparing the person in front with the image on the screen, whilst scanning the boarding pass, it will be akin to the LHR use of photographs in intra-terminal transfers e.g. to Northern Ireland in the old T2.
HKG works perfectly well with a barcode on the Passport, scanned at the booth, and a finger-print scan.
*A recent paper (
http://proceedings.mlr.press/v81/buolam ... ini18a.pdf) describes the use of three commercial facial recognition systems identifying
gender; the error rates for identifying the gender of white males was less than 1%, for darker skinned female subjects the error rate ran between 20 and 35%, for the darkest-skinned female faces, two of the tested systems had error rates of >40% when trying to identify gender.
