question

MohammedFazil-4342 avatar image
0 Votes"
MohammedFazil-4342 asked MohammedFazil-4342 commented

Face API - Facial Recognition-Microsoft Azure Cognitive Services Showing different confidence score

Face verification
Check the likelihood that two faces belong to the same person and receive a confidence score.
face


![109589-image.png][3]


Confidence return - 0.93


When we are comparing the two faces belong to the same person using the same Face API - Facial Recognition - Microsoft Azure Cognitive Services we got a different confidence score.

109599-image.png


[3]: /answers/storage/attachments/109641-image.png



Confidence return - 0.83


Please check the above result and please tell me what i have to do
i tried all the logic that i can use but did not get any same results

So experts please help me

azure-cognitive-servicesazure-face
image.png (561.8 KiB)
image.png (8.5 KiB)
· 1
5 |1600 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

@MohammedFazil-4342 Did you get a chance to review the below response and try to use the latest detection and recognition models?

1 Vote 1 ·

1 Answer

romungi-MSFT avatar image
0 Votes"
romungi-MSFT answered MohammedFazil-4342 commented

@MohammedFazil-4342 The result of the response depends on the detection and recognition models used for the images when they are uploaded using face detect. Usually the latest versions of these have a better accuracy i.e recognition_04 and detection_03 models and these models could be optimized further based on improvements to the backend algorithms.

The demo page above is displaying a sample which simplifies the scenario by using a certain set of parameters or values to create face ids using Face Detect and then verifying them using Face Verify. The score seen is also from this response. With the latest version of these models I receive a score of 0.94087 which is quite high and they are identical.

 {
   "isIdentical": true,
   "confidence": 0.94087
 }

I would recommend to use these models and not compare the score with demo site as this is periodically updated based on documentation changes. If you are using older detection and recognition models I would recommend to update the same for better accuracy. I hope this helps.


· 1
5 |1600 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

Thanks for the update.
we have updated the new request parameter (Recognition Model, Detection Model) in our code and now is returning a proper confidence score.

// Request parameters
queryString["returnFaceId"] = "true";
queryString["returnFaceLandmarks"] = "false";
queryString["returnFaceAttributes"] = "{string}";
queryString["recognitionModel"] = "recognition_04";
queryString["returnRecognitionModel"] = "false";
queryString["detectionModel"] = "detection_03";
queryString["faceIdTimeToLive"] = "86400";

0 Votes 0 ·