The database is likely pre-populated from the passport data, and many of the rules about how passport photos are taken are specifically designed so that they can get a good facial biometric from...
The database is likely pre-populated from the passport data, and many of the rules about how passport photos are taken are specifically designed so that they can get a good facial biometric from the image. It's been fairly standard for the last few years to go through fully automated customs & immigration in some countries: touch the passport to the reader, stand in front of the camera, and that's it.
Of course, when they read the passport chip first it means that the realtime photo can just be verified against that one person's data. I'm not particularly familiar with how facial recognition works mathematically, but I can say that a general lookup across 300m records of well indexed data isn't particularly onerous. Given that identifying an individual within a large population is a fairly common application of the technology, I would make an educated guess that researchers have put a lot of time into ensuring that the data can be stored and indexed in a way that makes this efficient.
In terms of reducing the search space (if necessary), I believe that the airlines are required to submit passenger manifests for all international flights entering the USA - I imagine a well designed lookup system would pull the data from those manifests and push it to the top of the facial recognition stack.
If the images are converted into a set of numerical characteristics (sizes/proportions) then a database lookup among 300 million records in 3-5 seconds doesn't seem that outrageous. I mean, Google...
If the images are converted into a set of numerical characteristics (sizes/proportions) then a database lookup among 300 million records in 3-5 seconds doesn't seem that outrageous. I mean, Google returns "About 237.000.000 results (0,62 seconds)" as a response to "facial recognition"...
I thought most facial recognition models use reference points which are not modified with facial muscle or jaw movement (e.g. distance between eyes, cheekbones, nose bridge, brow ridges, temples,...
I thought most facial recognition models use reference points which are not modified with facial muscle or jaw movement (e.g. distance between eyes, cheekbones, nose bridge, brow ridges, temples, etc), so contorting your face won't do you much good?
Do you wear glasses, by any chance? The automated machines didn't used to explicitly say to take them off, and I've had a 100% failure rate when wearing them but a near-100% success rate when I...
Do you wear glasses, by any chance? The automated machines didn't used to explicitly say to take them off, and I've had a 100% failure rate when wearing them but a near-100% success rate when I remove them.
Do they not have the pictures from passports on file or something?
The database is likely pre-populated from the passport data, and many of the rules about how passport photos are taken are specifically designed so that they can get a good facial biometric from the image. It's been fairly standard for the last few years to go through fully automated customs & immigration in some countries: touch the passport to the reader, stand in front of the camera, and that's it.
Of course, when they read the passport chip first it means that the realtime photo can just be verified against that one person's data. I'm not particularly familiar with how facial recognition works mathematically, but I can say that a general lookup across 300m records of well indexed data isn't particularly onerous. Given that identifying an individual within a large population is a fairly common application of the technology, I would make an educated guess that researchers have put a lot of time into ensuring that the data can be stored and indexed in a way that makes this efficient.
In terms of reducing the search space (if necessary), I believe that the airlines are required to submit passenger manifests for all international flights entering the USA - I imagine a well designed lookup system would pull the data from those manifests and push it to the top of the facial recognition stack.
If the images are converted into a set of numerical characteristics (sizes/proportions) then a database lookup among 300 million records in 3-5 seconds doesn't seem that outrageous. I mean, Google returns "About 237.000.000 results (0,62 seconds)" as a response to "facial recognition"...
I thought most facial recognition models use reference points which are not modified with facial muscle or jaw movement (e.g. distance between eyes, cheekbones, nose bridge, brow ridges, temples, etc), so contorting your face won't do you much good?
Do you wear glasses, by any chance? The automated machines didn't used to explicitly say to take them off, and I've had a 100% failure rate when wearing them but a near-100% success rate when I remove them.