Application of Spectral Information in Identification of Real-Fake Face Images
Akhilesh Kumar Pandey, Rajoo Pandey · 2015
The biometric authentication systems based on face recognition are easy to implement with any device, which has an in-built camera. These systems are very secure, but with various attacking methods, such security also becomes vulnerable. Among various types of possible attacks, spoofing is an attack in which available face information is presented before the sensor to mislead the authentication system. In this paper, a model has been presented based on spectral analysis of the captured images to classify them as real faces and face images. We consider Fourier and cosine transform of the image for evaluation of various Image Quality Measures (IQMs), with which the classification is performed using neural networks. The proposed spectral contents based IQMs are compared with several conventional IQMs to judge their performance. The simulation on Replay-Attack database has shown the improvement in the performance of the present model.