CounterfeitFace: A Convolution Neural Network Approach to Face Spoof Detection

Face anti-spoofing is a security feature that prevents unauthorized access to biometric systems using phony or synthetic facial photographs. It is a critical technology in mobile payments, access control, and law enforcement. This research offers a convolutional neural network (CNN)-based architecture for detecting face faking. The procedure starts with downsizing the input images, and then relevant frames are selected using silhouette score analysis and K-Means clustering. This maximizes efficiency by emphasizing frames with information. Face cropping separates relevant areas for detailed feature extraction. To distinguish between real and artificial faces, we extract texture information such as local ternary pattern (LTP) and local directional pattern (LDP) and image quality parameters such as brightness, chromatic moment, and blurriness. The CNN model is trained and tested using a Softmax layer to predict authenticity. Experiments on the NUAA and MSU-MFSD datasets reveal the CNN model’s superior performance, obtaining 98.31% and 98.0% accuracy rates, respectively, with lower attack presentation classification error rate (APCER), bona fide presentation classification error rate (BPCER), and half total error rate (HTER) values than other models.