-
Face Detection using Deep Neural Network (r-CNN) to create 128d embeddings of the face. Embeddings are created from 68 point landmarks of the face.
-
A simple SVM classifier is trained from the 128d embeddings alongwith the names.
-
Classifier used to recognize faces.
-
Eyes detected using landmark points. Blinking detected to prevent image spoofing by making calculations from the landmarks points.
-
Liveness v1 completed. Liveness detected by running OpenCV over different color channels (YCrCb and CIE LUV, others may hold potential too), and naive model trained to distinguish fake image/video of a face from real.