RNet-SGDG: An Improved Anti-Spoofing Time Efficient Framework for Face Recognition Using Deep Learning

Deep Learning Methods are efficiently used in image classification and computer vision these days like surveillance systems, gender prediction, defense, mobile applications, and face recognition. But due to the problem of different types of Spoofing attacks and the time to recognize a person, robust face recognition is still a challenging problem for researchers. This paper proposed a time efficient and anti-Spoofing face recognition which makes the system robust. Deep residual learning ReSNeTl01 is used to extract the deep features of face images. After the feature extraction, Classification is done in two steps. Firstly, the real spoof attack predictor is used to check the liveness of a person and after that subject-id is predicted in the second step. The stochastic Gradient Descent (SGD) classifier is used to classify the spoof or live person and the Gaussian Naivy Bayes classifier is used to recognize the person. The Replay Attack dataset is used for the experiment and achieves 99.724% accuracy, 99.687% f1-score, and 0.308 Seconds recognition time which is better than the existing techniques.