Scored
Blade outline
The silhouette of the blade, sampled around its boundary. It is quick to compare and rules out most non-matches, but it is weak on keys cut from the same blank, so it forms only part of the score.
How KeySafe matches a key.
Keys cut from the same blank can differ by a fraction of a millimetre, and a photograph taken at a different angle or distance can hide that difference. KeySafe takes six separate measurements from a scan and compares each one against the keys in your catalogue.
How a key is enrolled.
A key is photographed flat, from directly above, on both sides. Each photo is separated from its background so that only the key is stored. The two photos are then combined into a single record, which is what later scans are compared against.
Flat and square
Shooting from directly above keeps the outline and cuts true to shape, so scans compare like with like.
Both faces
The two faces of a key differ, and the reverse often carries the detail that separates two similar keys.
Millimetres where possible
A depth sensor records real dimensions, so the measurements do not change with camera distance.
The six measurements.
Each is taken separately from the others, so a poor frame affects one measurement rather than all of them.
Scored
Bitting profile
The cut of the teeth, measured as a thickness profile along the blade and reduced to a code. It carries more weight than any other single measurement.
Scored where available
Depth
On phones with a depth sensor, the key is recorded in three dimensions: width, height and thickness in millimetres, which do not change with how far away the camera was.
Scored, then gated
Bow shape and topology
The head of the key: its proportions, how round or square it is, how solid it is, and whether it has a cutout.
Gated
Dimples
The pockets milled into the faces of a key. Two keys that match on outline and bitting often differ in their dimples.
Gated
Warding ridges
The grooves running the length of the blade. These describe the keyway, which is the shaped opening the key passes through.
Codes and lettering stamped on the bow are also read, on devices that support text recognition.
How the measurements are combined.
The six measurements are not averaged together. Some of them can be approximately right, and some of them either match or they do not. The two kinds are handled differently.
Weighted
Bitting, outline and edge structure
These vary with the quality of the capture: a slightly blurred scan makes them slightly less certain rather than wrong. Each contributes a weighted score, and bitting carries the largest weight of the three.
Gated
Warding and dimples
Two keys with different keyways cannot be the same key, whatever the rest of the blade looks like. A warding or dimple mismatch reduces the whole score rather than being averaged in, which removes matches that a comparison of outlines alone would accept.
The result is a match score, along with the measurements it was built from. The app shows the score as a tier (Perfect, Close, Possible or Unlikely match) rather than a percentage.
What KeySafe cannot do.
These limits apply to every scan.
- It cannot identify a key it has never seen. Matching runs against the catalogue you built. There is no public database of key types behind it.
- It does not produce cutting codes. The bitting profile is used for matching, and is not shown in a form that could be used to cut a key.
- It does not name the blank. KeySafe does not identify the manufacturer or blank of a key, only which of your saved keys a scan matches.
- Shape processing runs in the cloud. Your library — browsing, searching, relabelling — works offline. Capturing a new scan and running a match need a connection.