spoofing detection with librealsense

[bingws]

Hi guys,

I'm in a project where I need active spoofing detection using depth information on librealsense sr300. Any idea about how to achieve this? Are there any simple demos? Unfortunately I can't use the RealSense SDK on this one :(

Thanks

[dorodnic]

Hi @bingws Spoofing detection falls into the category of higher level algorithms that you can implement using depth. The simplest approach is to try detecting flat planes, to rule out pictures of objects (I did a toy project like this some time ago, #1786) This is, however, super easy to break. A better solution would be to gather a lot of "real" data (say human faces) and a lot of "fake" data (pictures, but also masks, 3D prints etc...) and train a classifier to distinguish between the two.

[MartyG-RealSense]

@dorodnic bingws also commented on a discussion about recognition of real / fake images that was created by another user, where you suggested using a Haar Cascade in OpenCV and deep-learning classifier training. I'll post a link to the answer I provided there so that others can easily find the information.

https://github.com/IntelRealSense/librealsense/issues/2133#issuecomment-471426787

[bingws]

@MartyG-RealSense I've learned about face recognition in 2D before. Haar Cascade seems to be a face detection model without depth information, which is easy to crack through photos. @dorodnic "rs-antispoof" is out of date. Could you send me a complete copy of "rs-antispoof.cpp"? I'll study. Thank you.

[MartyG-RealSense]

An alternative way to train a system to tell the difference between real and fake images may be to use the TensorFlow software.

https://www.tensorflow.org/

It has a tutorial on this subject.

https://www.tensorflow.org/alpha/tutorials/generative/dcgan

[bingws]

@MartyG-RealSense This tutorial is not about detecting a living body. Although I don't understand this example, it seems to test the authenticity of the photo.

[MartyG-RealSense]

There is a tutorial on medical image classifying in this YouTube link:

https://www.youtube.com/watch?v=DCcmFXXAHf4

It has an accompanying GitHub for use with TensorFlow.

https://github.com/llSourcell/AI_in_Medicine_Clinical_Imaging_Classification

[bingws]

@MartyG-RealSense The following example is not available.Can you send a full version?Thank you.

[MartyG-RealSense]

It looks as though all the code relating to antispoof can be found at the link below.

https://github.com/IntelRealSense/librealsense/pull/1786/files

With the rs-antispoof.cpp section specifically at this link:

https://github.com/IntelRealSense/librealsense/pull/1786/files#diff-ec49a1383a83b87bf03f5c50e1b8da3c