Enable Face Liveness Detection

The FindFace Multi face liveness detector tells apart live faces from face representations, such as face images, videos, or masks. The liveness detector estimates face liveness with a certain level of confidence and returns the confidence score along with a binary result real/fake, depending on the pre-defined liveness threshold.

On video streams from cameras and uploaded video files, liveness detection is performed by the findface-video-worker service. You can enable liveness detection during FindFace Multi standard installation by answering y to the question Enable liveness and attempt to continue installation?(y/n). If you skipped this step, you can manually enable liveness later, following the instructions below.

Note

The face liveness detector functions on both CPU- and GPU-acceleration. However, it is much slower on CPU.

In addition to liveness detection, realized by the findface-video-worker service, FindFace Multi provides an API-based face liveness detection service findface-liveness-api. The findface-liveness-api service is automatically installed and enabled during the standard deployment, as FindFace Multi uses it for authentication by face. The service supports CPU- or GPU-decoding.

Liveness detection can be enabled on external detectors. Frames from external detectors are processed by the findface-extraction-api service. The service supports CPU- or GPU-acceleration. The liveness extraction model faceattr.liveness_web.v1, that comes in the default findface-extraction-api.yaml configuration, is used for authentication by face. If your use case implies liveness detection on an external detector, you may want to replace the default liveness recognition model in the findface-extraction-api.yaml file with the model that is recommended for liveness spoofing attack detection on PACS cameras. Read more in the Configure Liveness Recognition Model section.

In this section:

Enable Face Liveness Detector

To enable the face liveness detector on video streams and uploaded video files, do the following:

1. Open the /opt/findface-multi/configs/findface-video-worker/findface-video-worker.yaml configuration file. In the liveness section, specify the neural network models as shown in the example:

   GPU

   sudo vi /opt/findface-multi/configs/findface-video-worker/findface-video-worker.yaml
   
   liveness:
     fnk: /usr/share/findface-data/models/faceattr/faceattr.liveness_pacs.v3.gpu.fnk
     norm: /usr/share/findface-data/models/facenorm/crop2x.v2_maxsize400.gpu.fnk
   ...
   

   CPU

   sudo vi /opt/findface-multi/configs/findface-video-worker/findface-video-worker.yaml
   
   liveness:
     fnk: /usr/share/findface-data/models/faceattr/faceattr.liveness_pacs.v3.cpu.fnk
     norm: /usr/share/findface-data/models/facenorm/crop2x.v2_maxsize400.cpu.fnk
   ...
   

2. Restart the findface-multi-findface-video-worker-1 container.

   sudo docker restart findface-multi-findface-video-worker-1
   

Configure Liveness Threshold

If necessary, you can adjust the liveness threshold in the /opt/findface-multi/configs/findface-multi-legacy/findface-multi-legacy.py configuration file. The liveness detector will estimate face liveness with a certain level of confidence. Depending on the threshold value, it will return a binary result real or fake.

Note

The default value is optimal. Before changing the threshold, we recommend that you seek advice from our experts at support@ntechlab.com.

sudo vi /opt/findface-multi/configs/findface-multi-legacy/findface-multi-legacy.py

FFSECURITY = {
    ...
    # feature specific confidence thresholds
    'LIVENESS_THRESHOLD': 0.885,  # model: [faceattr.liveness_pacs.v3]
    ...

Restart all FindFace Multi containers.

cd /opt/findface-multi/

sudo docker-compose restart