Liveness detection - other than pulse

[idata]

So far, I've found that when using facetracking even with color plus depth, or even on IR mode.

If I hold a photo of a face up, it detects it as a face. Other than by using pulse. Is there any way for me to detect a real face from a spoof?

Thanks!

[idata]

I would like to know exactly the Realsense camera model number and the Realsense software version you are currently running on the system.

Allan.

[idata]

Model No: VF0810, RealSense SR300

SDK 2016 R2, version: 10.0.26.0396

Thank you for looking into this!

[idata]

Thanks for the product information.

I would like to replicate this problem and I will need detail description of the issue; background history, operating system version.Allan.

[idata]

Using Windows 10,

If you actually just run the Intel Realsense sample browser sample code "Face Tracking (C# )"

and in any of the tracking modes set, if you hold a photo up to the camera. You will see that it detects it as a face.

I am implementing a similar system in my code for the face tracking.

Even when setting the trackingtype to FACE_MODE_COLOR_PLUS_DEPTH(3D tracking) or FACE_MODE_IR(IR tracking), It still detects the faces in photos.

Is there any way that you suggest or any way that the sdk offers for me to implement face detection so that faces in photos are NOT detected (for security reasons and spoof detection)

Is there any implementation in the SDK that relies more on perhaps the depth information to not detect photo faces since the COLORPLUSDEPTH option does detect photo faces?

The pulse detection requires the user being in front of the camera for too long.

If I bend a photo in various ways I can sometimes trigger the code to detect that the eyes closed/mouth opened etc, so making my code sense facial expression changes isn't a very good liveness detection method either.

My best idea is for the detection to rely on depth or perhaps there is some method for doing this that I've missed that the realsense camera can implement?