Sending 4k screenshots twice a second to a server would be tremendously bandwidth hungry. My guess is that it's all done locally.

▲

treyd 17 hours ago | parent [-]

There's probably compact signatures extracted from the screenshots (color profiles, OCR, etc) which are then uploaded later in bulk. You don't need the full original image to be able to reliably uniquely identify the content if you have an index of it already.

▲

floxy 17 hours ago | parent [-]

I'm wondering if there is some sort of steganographic watermark that broadcasters are including in media, to enable stuff like this. Probably would need to be robust in the presence of re-encoding, more compression, etc..

	▲	inetknght 16 hours ago \| parent [-]
		This has been long solved by YouTube for detecting CP and other non-compliant videos. For example, check out https://github.com/akamhy/videohash