Why PhotoGraph
Facial recognition matches faces.PhotoGraph finds people.
Facial recognition answers one narrow question about two clear faces. PhotoGraph finds every image of a person, with the situational context around it, and can identify every other person in each image.
Side by side
A different question needs a different engine.
| Facial recognition | PhotoGraph | |
|---|---|---|
| Question it answers | Is this face a match for that face? | Is this person anywhere in this collection? |
| What it needs | A clear, front-facing, well-lit face | Any photo of the person, including partial, shadowed or turned away |
| Across a lifetime | Accuracy drops outside a tight age range | Groups a person across their whole lifespan as one identity |
| Group and crowd photos | Misses hidden people, or everyone when faces are too small | Sees every face in the photo to provide a match |
| When there is no face | Can’t operate | Uses context, transitivity and a patent-pending method to find anyone, anywhere |
| Setup | Often tuned or trained on labeled examples | No labels, no training, pixels only |
| Output | A match score per pair of faces | A queryable identity graph for the whole collection |
| Where it runs | Frequently a cloud API | Your environment, on your hardware, protecting your valued assets |
The performance gap
Nothing on the market came close, so we built a real-world test.
Public benchmarks use clean, recent, front-facing photos. Real collections don’t look like that. So we tested PhotoGraph using a true family collection of 33,000 photos and one known person who appears in 524 of them: every age, film scans to digital, alone and in groups.
The question was simple. Which method will find the most photos of the correct person? The correct answer was all 524.
Every photo a method misses is content lost: invisible to a subscriber searching for family, and left for an analyst to find by hand.
Which method will find the most photos of this person?
The real-world test: a true 33,000-photo family collection. One known person appears in 524 of them, ages 2 to 76, film scans to digital, solo and in groups.
For perspective: with other methods, you see only a fraction of this person’s photos, and miss what could be the heart of the narrative somewhere among the remaining ~33,000 images. With PhotoGraph, the job is done.
What makes the difference
Four reasons buyers choose PhotoGraph.
Pixel-based entity resolution, not matching or text linking
You gain a new level of integrated entity resolution that enhances current operations and opens new revenue.
Works out of the box
No data collection, labeling or training. Results on your own collection from the first deployment.
Sovereign deployment
Runs inside your environment. Available where cloud AI services are ruled out by law, regulation or policy.
Lean compute & speed
PhotoGraph doesn’t require training, so compute and time-to-market are vastly more efficient.
Run the test on your collection.
If you’re a business with a large photo archive, we can run a pilot on all or a subset of your data. Or, if you want to leap right in, let us give you a briefing and discuss scoping.