0

I've been working on a facial detection and recognition program for a few days now in OpenCV using Eigen/Fisher/LBPH FaceRecognizers that will compare the faces in two photos using the 3 listed recognizers and return a confidence value that the faces are the same person or not.

While I've been able to get everything working, the results and recognition rates have not been inspiring, especially when you look at a service like Microsoft Face API (which I cannot use due to privacy concerns) at this url: https://azure.microsoft.com/en-ca/services/cognitive-services/face/

Does anyone here have any idea what method(s) Microsoft is using in their Face verification on the above URL? It's exactly what I need (my tests have shown it to be extremely accurate for my scenario), aside from the fact that it's an API and not an SDK.

Scott Dellinger
  • 158
  • 1
  • 12
  • Conventional methods like eigen/fisher etc only takes you to some level. If you want accuracy beyond that try Deeplearning. Check out https://github.com/davidsandberg/facenet. But the FPS of facenet is less than that eigen/fisher faces achieve. – harshkn Aug 17 '17 at 07:16
  • Microsoft (as all competitors) funds heavy research on this for many years. I suppose this is all surrounded by trade secrets. You can focus on the people and their publication though, for example public publications from this https://www.microsoft.com/en-us/research/people/doch/ could give some hints. – Simon Mourier Aug 17 '17 at 08:17
  • Microsoft also published this a few years ago: https://www.microsoft.com/en-us/research/project/microsoft-research-face-sdk-beta/ you can still download it. – Simon Mourier Aug 17 '17 at 08:23

0 Answers0