4 ms·
It is for facial recognition software. I know nothing about the field, but my assumption would be that you need a plain base to work from. That way you start
by brc 13y ago
It is for facial recognition software.
I know nothing about the field, but my assumption would be that you need a plain base to work from. That way you start with a known value and you compare from the base to the image you're trying to match.
If you had a smiling photo as the base, and you're trying to match a photo where the person is frowning, that must be that much harder to work out.
Remember that smiling changes not on the mouth but the eyes as well. I would guess that eye shape is a key factor in making facial recognition work.
For the record, I am favor of facial recognition for identity checks with passports. It means I can use the automated immigration processing queues, which are much, much faster than standing in a line waiting for a guy to ask you a couple of benign questions and stamp your passport.
- therobot24 13y agoEye shape is good in NIR images, otherwise in VL images it not as big a factor.