On Mon, Mar 05, 2018 at 06:12:37PM -0800, Christian Huitema wrote: > Required resolution depends on context. Face recognition works by > comparing the captured face to a set of N possible candidates, and > returning the best match. The more pixels you have, the more precise the > recognition, but there are claims that as little as 64x64 can be used > for recognition -- probably for finding one user among 4096. Basically, > if you can recognize the face in the picture, you have to assume that > machines can do that too. Well, i will stay optimistic, also in face of some good examples i have seen how bad recognition still is. Not only thinking of gorillas. And of course to your data points because you'd most likely have to try to find a face in millions in real use cases. > Reusing the same picture everywhere seems like a reasonable strategy. Of > course, that will not stop the immigration agents who are taking your > picture while scanning your passport. Sure, but whats published on the Internet is likely more controllable by what i do freely. > Someone mentioned paper bags, but people are seriously looking at makeup > and accessories specifically designed to fool the algorithms. Check for > example https://cvdazzle.com/. Maybe that will be the new IETF look... I forgot the name of the system, but back in the early 90th we had some database of face pictures, which i think we used also on mbone or the like, and i am sure Van Jacobson hat a picture with a paper bag over his head... If i look for my name on google picture search, i only find the few picture that have my name in some metadata, so i have not really seen publically any good example of face recognition. Are there any good examples publically available ? Cheers Toerless
