Similar Faces and Merging
Duplicate subjects happen. Somebody is enrolled twice under different spellings, or a visitor is enrolled again on a later visit. The Similar Faces panel finds them, and merging resolves them.
New in version 11: merging is carried out by the server as a single orchestrated operation, so a merge cannot be left half-applied.
Finding duplicates
The Similar Faces panel on a subject profile shows other subjects whose photographs closely resemble this one. Each is a candidate for being the same person.
Reviewing it is worth doing whenever you open a subject you are about to act on. A person split across two records has their detection history split too, so both records understate how often they have been seen — which is exactly the figure you were about to rely on.
Candidates appear as a grid of portrait cards. Each carries badges for what is known about it — notes, shared notes, tags, alert tags, whether alerting is enabled, and whether the subject is un-enrolled — so you can judge a candidate without opening it. Hover a note badge to read the note.
Select all and Clear work on the whole grid, and the footer says exactly what will happen: Merge N selected into this profile.
Where a card is not enough, the name opens that subject's profile, and the icon beside it opens it in a window of its own — which is how two profiles get compared side by side rather than from memory.
Deciding whether to merge
Look at more than the photographs:
| Evidence | Suggests |
|---|---|
| Same name or ID | Same person |
| Detection histories that interleave | Same person |
| Detections at the same moment in different places | Different people |
| Different employers or departments | Probably different |
The third is the strongest negative signal available, and worth checking deliberately.
Merging brings two records together into one. Where they were genuinely different people, their detections, tags and history are now conflated — and the record no longer distinguishes them.
Check before merging, not after. Two records are an inconvenience; one wrong record is a wrong answer to every future question.
Merging
Choose the merge from the Similar Faces panel and resolve any conflicts the dialog presents — where the two records disagree on a field, you choose which value survives.
The merge is applied server-side as one operation, so it either completes or does not. A subject cannot be left partly merged.
Splitting
The opposite case: photographs of two different people have ended up on one subject, usually from a mistaken enrolment. Splitting separates them into distinct records.
Both are audited — types 1008 and 1009 — so a merge or split that turns out to be wrong can be traced to who performed it.
Merging by name in bulk
One profile at a time is the right approach when you are investigating. When you are tidying up — after an import, or after a spell of enrolment by several people — Identity Merge ▸ Merge By Name under the Face & LPR settings works through the whole system at once.
It finds every group of known subjects sharing the same first, middle and last name and lists the groups down the side. Pick a group and its profiles appear as cards with the same badges as above, plus a detection count. Select the ones that are genuinely the same person — the first one you select is the one that is kept — and merge.
Auto Merge goes further: it face-compares the profiles within each group and merges those the comparison agrees on, leaving the rest for you. It works from the enrolled faces, so a profile with no usable face is left alone rather than merged on its name.
Use it on a set you have already spot-checked, and read the result before moving on; shared names are common, and two different people called the same thing is exactly the case the name grouping cannot resolve on its own. The confirmation says what is about to happen and which profile is kept — read it rather than clicking through, because a merge is far easier to avoid than to unpick.
This surface needs the bulk enrolment rights, and appears only where face recognition is enabled.
Preventing duplicates
Cheaper than resolving them:
- Search before enrolling. A face search first shows whether the person is already known.
- Read the close-match warning during enrolment rather than clicking past it.
- Agree a naming convention. Most duplicates begin as two spellings of one name.
Audit record
What eConnect writes to the audit trail for the actions on this page.
FACE SUMMARY PULLED FOR FACE '{subject name}'