Enhanced Face Liveness Detection Based on Features From Nonlinear Diffusion Using Specialized Deep Convolution Network And Its Application In OAuth
Loading...
Authors
Alotaibi, Aziz
Issue Date
2016-12-30
Type
Thesis
Language
en_US
Keywords
Facial detection , Biometric authentication , Open Authorization
Alternative Title
Abstract
The major contribution of this research is the development of enhanced algorithms that will prevent face spoofing attacks by utilizing a single image captured from a 2-D printed image or a recorded video. We first apply a nonlinear diffusion based on an additive operator splitting (AOS) scheme with a large time step to acquire a diffused image. The AOS-based scheme enables fast diffusion that successfully reveals the depth information and surface texture in the input image. Then a specialized deep convolution neural network is developed that can extract the discriminative and high-level features of the input diffused image to differentiate between a fake face and a real face. Our proposed method yields higher accuracy as compared to the previously implemented state-of-the-art methods. As an application of the face liveness detection, we develop face biometric authentication in an Open Authorization (OAuth) framework for controlling secure access to web resources. We implement a complete face verification system that consists of face liveness detection followed by face authentication that uses Local Binary Pattern as features for face recognition. The entire face authentication process consists of four services: an image registration service, a face liveness detection service, a verification service, and an access token service for use in OAuth.
Description
Citation
A. Alotaibi, "Enhanced Face Liveness Detection Based on Features From Nonlinear Diffusion Using Specialized Deep Convolution Network And Its Application In OAuth", Ph.D. dissertation, Dept. of Computer Science and Engineering, Univ. of Bridgeport, Bridgeport, CT, 2016.