Are the trained face images retained in the database?

[sybersitizen]

The images used to train faces come from user files that IMatch references. Are those trained face images retained in the database, independent of the user files that they came from? In other words, are there any consequences if the files from which the trained images originated are later removed from the database or deleted?

[Mario]

IMatch creates fingerprints from each face is recognizes. Basically a mathematical description of the face features. This are stored in the database. If you remove the file containing the face annotation, all linked information is removed, including the fingerprints of the faces in that image. Removing a file also removes the link from that file to the person(s) linked to the image, among other things.

IMatch stores the cover file as an image in the database. Removing the file containing the face annotation from which the cover file of a person was created has no impact on the cover image.

[sybersitizen]

Then it sounds like removing or deleting *all* of the files that have been used to generate trained face data for a person would destroy the functionality of face recognition for that person. True?

But is it correct that removing or deleting *some* of the files that have been used to generate trained face data for a person is not a problem as long as enough other files that were used for that purpose remain in place?

[Mario]

Basically, up to a point, the more face fingerprints of a "person" are available in the database, the better the recognition rate will be. On the other hand, the more different faces of a person are available, the more unspecific face recognition rates will become. Having day of birth data available for a person will improve recognition rate. Having many unconfirmed faces reduces recognition rate. It's all a play of tensors, in the end.