Match probability reporting allows operators to distinguish between exact matches and probable matches, enabling quicker responses to emerging situations. This information is only reported when using the Proximity Threshold Allowance (in Recognition Settings). When the actual difference between the subject and the reference image is between the Identity Recognition Threshold and the Proximity Threshold Allowance, the subject is reported as a “Close Match” as shown below in the SAFR Events Viewer.
You can simulate this by adjusting the Identity Recognition Threshold well below the default value of 0.54 and setting Proximity Threshold Allowance to something high like 0.8. (Remember to restore the defaults as these settings result in high rates of false negatives.)
The probability value reported will change as SAFR confidence increases and will not be displayed when the subject match reaches 100% confidence.
Note: Face recognition is always a measure of how different the subject is from the reference image. 100% confidence should be interpreted as the subject and the reference image differing by the configured value of Identity Recognition Threshold. The following provides guidance on how to interpret confidence when using default values for Identity Recognition Threshold.
- >= 100% Certain match (values > 100% are possible which indicate even greater certainty)
- 93-100% Close match but not certain enough to unlock the door
- 86-93% Possible match with low confidence
- 82-86% Similar face with no confidence of match
- 0-82% Different face
The same information is also reported as “confidence” in SDKs and Events Service REST API. More information on the SDK is available here <https://docs.real.com/sdkdocs/SDK_Docs.html> and information about SAFR REST APIs can be found here <https://cv-event.real.com/docs/index.html>