Generalized Face Spoofing Detection Using Spatial Features

Authenticating a person using biometric traits like fingerprint, iris, and face is widely adopted in security applications. Compared to other authentication procedures biometrics provides high accuracy. Face biometrics possess high precision and low intrusiveness. However, face recognition-based security systems are vulnerable to face spoofing attacks. Using printed photo or video replay intruders attempt to break the recognition system. Face spoof detection systems classify real and fake users by analyzing the features. Various spoof detection techniques have evolved based on texture and quality features. However, these methods suffer from generalization ability and variations in textures. So, in this research work, a face spoofing detection model is presented that utilizes spatial and frequency features to detect real and fake users. Feature descriptors like Histogram of Gradient (HOG) and Gray-Level Co-occurrence Matrix (GLCM) are used to obtain the essential features and extracted features are processed through a random forest classifier module. Extensive experiments on benchmark datasets namely the Reply-attack database and MSU-Mobile face spoof database demonstrate the better detection performance of the proposed spoof detection model compared to state of art of techniques. Cross-database evaluation of the proposed approach possesses minimum error indicates the generalization ability of the proposed face spoof detection model.