I want to get an expert opinion. What is the easiest way to compare passe-partout images for uniqueness?

Question

Image processing

I have a long-standing pain (to remove the routine). Non-commercial project in which there are performers. To register, the performer uploads his name, passport photo and a short video about himself.
The main problem is that there are performers who violate the rules of work and get banned.
But it is not a problem for them to take another phone number and register again in the service and continue to mow, while they use the same passport.
At the moment (for 3 months) more than 18 thousand registrations. The percentage of successful ones is very small (no more than 40-50%). But there are repeated registrations with the same passport.
For a long time I want to figure out how to analyze the series / passport number and check them for uniqueness.
If you enter this data manually, I’ll just go nuts and get bogged down in a routine (because I’m not only doing this).
I will say right away that I went to Google and even beyond the 10th page. And not only so I tried to find the answer to my question.
I would like to hear the opinion of experts and get advice, maybe someone has already encountered such problems. Reading numbers from the image and comparing the received with the base.
I would like to hear your recipes and evaluate whether I can implement with my resources.
Thank you all in advance!

Answer 1

https://www.pyimagesearch.com/2015/11/30/detecting...
in fintech they actually hire people, it is checked and entered manually, the main problem is that
SIFT-SURF-ORB numbers/data are faked, they don’t work very well, precisely because for the sameness of the document, strictly speaking, the photo only saves,
so you need to demand at least 600px,
but with a video story, everything is simplified by the
way, recently there was an article on Habré about automating sales of SIM cards