Face antispoofing based on frame difference and multilevel representation

Abstract

Mourad Oussalah, "Face antispoofing based on frame difference and multilevel representation,"Abstract. Due to advances in technology, today's biometric systems become vulnerable to spoof attacks made by fake faces. These attacks occur when an intruder attempts to fool an established face-based recognition system by presenting a fake face (e.g., print photo or replay attacks) in front of the camera instead of the intruder's genuine face. For this purpose, face antispoofing has become a hot topic in face analysis literature, where several applications with antispoofing task have emerged recently. We propose a solution for distinguishing between real faces and fake ones. Our approach is based on extracting features from the difference between successive frames instead of individual frames. We also used a multilevel representation that divides the frame difference into multiple multiblocks. Different texture descriptors (local binary patterns, local phase quantization, and binarized statistical image features) have then been applied to each block. After the feature extraction step, a Fisher score is applied to sort the features in ascending order according to the associated weights. Finally, a support vector machine is used to differentiate between real and fake faces. We tested our approach on three publicly available databases: CASIA Face Antispoofing database, Replay-Attack database, and MSU Mobile Face Spoofing database. The proposed approach outperforms the other state-of-the-art methods in different media and quality metrics.

Face Spoof Attack Detection using Deep Background Subtraction

“…Despite the fact that our EER and HTER are similar to the previous approaches. The following are the types of attacks on replay databases that we compute the performance of using our method: Depending on the method used to hold the 11.78 11.79 LBP-TOP [11] 07.90 07.60 IDA [13] 08.58 07.41 Motion+LBP [49] 04.50 05.11 FD-ML-LPQ-Fisher [6] 05.62 04.80 DMD [12] 05.30 03.75 Colour-LBP [15] 00.40 02.90 Spectral cubes [41] -02.75 CNN [20] 06.10 02.10 USDAN-Norm [22] -00.30 Bottleneck Feature Fusion + NN [23] 00.83 00.00 Identity-DS [21] 00.20 00.00 S-CNN+PL+TC [46] 0.36 -BS-CNN+MV (our) 00.58 00.62 attack replay device (paper, mobile phone, or tablet), the three attack subsets (print, mobile, and highdef) were recorded in two different modes: i) fixed-support and ii) hand-based (See Tables 3). We also put our method to the test using the MSU-MFSD database.…”

“…Inspired by the work of Frame Difference and Multilevel Representation (FDML) [6], we propose effective biometrics systems based on the detection of face spoofing. To do this, we suggest using the background substruction method in the preprocessing step to adjust the face's motion.…”

Spoof detection using Elegant Pairing method on Stockwell Transform in Face Biometric System

“…Benlamoudi etal. [2] combine three popular texture descriptors, BSIF, LBP, and LPQ. Thippeswamy,G et al [38] created an ensemble of Local appearance based models based on LBP, GDP, GLTP, LDiP, LGBPHS and LPQ descriptors to detect 2D presentation attacks.…”

Attention-Based Two-Stream Convolutional Networks for Face Spoofing Detection

“…Due to the domain shift (different imaging environments) between databases, the performaence of all the anti-spoofing methods drops. Compared with the state-of-the-art methods, [60] 50.2% 47.9% LBP [56] 55.9% 57.6% LBP-TOP [61] 49.7% 60.6% Motion-Mag [64] 50.1% 47.0% Spectral cubes [22] 34.4% 45.5% CNN [14] 48.5% 39.6% Color-LBP [10] 47.0% 39.6% Colour Texture [8] 30.3% 37.7% Depth + rPPG [63] 27.6% 28.4% Deep-Learning [13] 48.2% 45.4% KSA [65] 33.1% 32.1% Frame difference [66] 50 our method (MobileNet + Attention) achieves the 2nd best performance (30.0% and 33.4%), slightly worse than the best one [63] (27.6% and 28.4%). However, [63] uses more auxiliary information (3D face shape, rPPG signals) than our method.…”