Face Spoofing and Presentation Attack Detection

Despite of not being much secured, face recognition has become the most popular security service medium. The inevitable result was a threat of fraud caused by face spoofing and Presentation Attack detection which is achieved by using images or videos of the users. To stop this, this paper suggests a method based on multi task cascaded convolutional neural network (MTCNN), Facenet and Support Vector Machine (SVM) classifier. This method detects the faces in the images using MTCNN, extracts the feature vectors using FaceNet and ultimately classify the images to be real or spoofed based on a SVM Classifier. This method is applied on the NUAA public anti face spoofing dataset and has achieved 79% accuracy on an average on test dataset. However using bagging or boosting technique with tree classifier in case of classification and training the model over bigger datasets like Celeb-A can improve the accuracy of the models’ prediction.