In Facial Recognition, What Are Probes and Galleries?

Probes and galleries aren’t necessarily diabolicial medical instruments or trendy art exhibits. They are key to facial recognition.

I started out in the fingerprint world before entering the facial recognition world. To a certain extent, some fingerprint knowledge is transferable to faces. You process images and find things in the images. Then you compare what you find, either manually via expert or by automated computation…or both. Finally, use this comparison to determine the probability of whether the two images come from the same person.

But there are limits to this transferability. If you ask a forensic face expert to locate the minutiae in a face, they will know that you are a dunderhead. And if you talk about bifurcations, ridge endings, and lakes, forget it.

By the same token, certain terms used in facial recognition are not explicitly found in the fingerprint world. I’ll look at two of them: probe and gallery.

Probe

Now I could go ahead and make my own definition up, but I prefer to use an authoritative one. So I’m turning to the OSAC Lexicon, developed by the Organization of Scientific Area Committees for Forensic Science. Of course Michele Triplett developed her Fingerprint Dictionary long ago, but it, um, targets fingers rather than faces.

So what is OSAC’s definition of a probe?

“in facial identification, a facial image or template searched against the gallery in a facial recognition (FR) system.”

This is what fingerprint people refer to as the search image. For example, let’s say that there’s video of someone robbing a 7-Eleven. The frame from the video camera that captures the robber could (if of sufficient quality) become the probe.

Probes and galleries, probe. Google Gemini.
Probes and galleries, probe. Google Gemini.

Gallery

In a one-to-many (1:N) search, there is a database of faces against which you search the probe. This is called a gallery. Or, as OSAC words it:

“in facial identification, a FR system’s database, which typically contains all known-person biometric references (samples or templates, or both).”

Using our 7-Eleven example, we can take that video frame and compare it against a database of all known criminals. This would be the gallery.

Probes and galleries, gallery. Google Gemini.
Probes and galleries, gallery. Google Gemini.

In essence, this is the computerized equivalent of a police lineup. Except that when you use a computer, that gallery may have many more than 6 people. In India, it could have over a billion.

So those are my facial recognition terms for the day. Let me ask you: should I explore more facial recognition terms? Or move on to iris? Or go back to fingerprint?

Postscript

See my Facebook short on the “starting lineup.”

It’s the starting lineup. Jelify.

Leave a Comment