Deep Discriminative Restricted Boltzmann Machine (DDRBM) for Robust Face Spoofing Detection

Biometrics emerged as a robust solution for security systems. Despite, nowadays criminals are developing techniques to accurately simulate biometric traits of valid users, process known as spoofing attack, in order to circumvent the biometric applications. Face is among the main biometric characteristics, being extremely convenient for users given its non-intrusive capture by means of digital cameras. However, face recognition systems are the ones that most suffer with spoofing attacks since such cameras, in general, can be easily fooled with common printed photographs. In this sense, countermeasure techniques should be developed and integrated to the traditional face recognition systems in order to prevent such frauds. Among the main neural networks for face spoofing detection is the discriminative Restricted Boltzmann Machine (RBM) which, besides of efficiency, achieves great results in attack detection by learning the distributions of real and fake facial images. However, it is known that deeper neural networks present better accuracy results in many tasks. In this context, we propose a novel model called Deep Discriminative Restricted Boltzmann Machine (DDRBM) applied to face spoofing detection. Results on the NUAA dataset show a significative improvement in performance when compared to the accuracy rates of a traditional discriminative RBM on attack detection.