Face anti-spoofing detection based on DWT-LBP-DCT features
Introduction
In recent years, the prosperous development and widespread application of intelligent technologies have placed more stringent requirements on the safety of biometric authentication systems. Different biometric authentication systems used different physical or behavioral characteristics to identify persons [1]. Alternatives have been explored over the years, including iris [2], voice [3], fingerprint [4], face information [5], and handwriting signatures [6]. Among them, face information becomes one of the best choices for system authentication from the perspective of low recognition cost and convenience. At present, the maturity of identification technology has enabled the biometric authentication system to be applied in many occasions, ranging from access control systems for various secret occasions to landing systems for mobile terminals, and even to the face recognition technology in mobile payment systems [7]. Face biometrics have achieved remarkable performance over the past decades, but unexpected spoofing poses a threat to information security [8].
Compared with other biological features (such as iris and fingerprint), pictures and videos of legal users can be easily obtained through social networks. The low-cost characteristics make fraud behaviors for face recognition especially easy to be implemented. Illegal users can imitate the effective biometrics of a legal user through the camera. Therefore, the robust face anti-spoofing algorithm is particularly important in face recognition technology. Only when the face is judged to be living and belongs to a legal user, recognition result is true and effective; otherwise, this is determined as an illegal attack on the authentication system [7]. Under this double guarantees, the security and reliability of the system can be improved. As a result, improving the anti-spoofing ability of the face recognition system has become an urgent problem in face authentication [7]. The effective face algorithm is not only important for system security, but also has research value and broad prospects.
In general, face attacks can be divided into three categories: attacks by photos, attacks by videos, attacks by masks. Photo attacks mean an illegal user prints the photo as a paper version or displays it on an electronic device, and presents the photo to the camera of verification system. A video attack is that an illegal user replays a video belonging to a legitimate user and attacks the face recognition system with dynamic information. Mask attacks refer to an illegal user wearing the 3D mask of the original user, imitating the stereo effect of the face. In attacks by photos and videos, there is a possibility of artificial modification, that is, the background in the picture may be the original background, but the foreground image (face area) is generated by software simulation or manually replaced. The APP software ZAO which replaces face in videos with another face provided by users, demonstrates that face anti-spoofing technology has gradually matured and there are also various potential unknown forms of attack in the future. Considering all of these, we put our scope on photos and video attacks in this paper, and focus on a generalizable method.
In terms of the attacks, traditional hand-engineered features have been proposed to describe liveness. Texture characterization was often envisaged by measuring the fluctuations (with respect to space) of image amplitude regularity [9], and made great results like Local Binary Pattern (LBP) and Dynamic Mode Decomposition (DMD). In the meantime, with the improvement of computer computing performance and the development of deep learning techniques, the performance of deep learning methods are gradually catching up with the traditional methods in the detection.
Compared to previous work, the contribution of this paper can be summarized as:
- •
Discrete Wavelet Transform (DWT) is utilized to present multi-resolution and location information by dividing image into coarse-scales and fine-scales.
- •
Local Binary Patterns (LBP) features is performed on the DWT features to identify the texture and amplify the difference between true/fake frame blocks.
- •
Discrete Cosine Transform (DCT) is applied on the DWT-LBP features to obtain main energy components of a video.
- •
A SVM classifier is trained with the DWT-LBP-DCT features and tested by using intra-test and combined-test strategies in two benchmark databases. Experiment results have shown that the proposed method can provide higher detection accuracy.
The rest of the paper is organized as follows. We briefly review the state-of-art literatures of face anti-spoofing detection in Section 2 and explain the proposed approach in detail in Section 3. Then, two databases and evaluation protocols are introduced in Section 4. This is followed by a report of experimental results in Section 5. Finally, conclusions and future work are given in Section 6.
Section snippets
Related work
Feature extraction and classifier designing are two key steps for reliable detection in images. Reviewing the relevant research, the methods mainly used for extracting features for face attacks can be divided into five parts: static texture based, motion based, frequency based, color based and deep learning based. And the majority of traditional face detection methods combines manual features and a simple machine learning method to construct an anti-spoofing system.
Static texture based:
Proposed method
In the paper, we present a DWT-LBP-DCT features based approach which has combined the characteristics of DWT, LBP and DCT operations to form a high-level feature for face anti-spoofing attacks. The flowchart is shown in Fig. 1. The pipeline is to extract frames at first. After face recognition and area enlargement operation, we have performed a DWT operation on each frame. All input and output images for DWT are RGB images, which are converted to gray versions for LBP operations. After a DWT
Face anti-spoofing databases
In order to evaluate the performance of the proposed face anti-spoofing algorithm, we adopt two benchmarking databases, which are CASIA-FASD database and REPLAY-ATTACK database.
Experimental results
In this section, we tested the performance of the proposed DWT-LBP-DCT method by using two standard benchmark databases. Also, we analyzed functions of the DCT and DWT operations in the proposed method by applying the DWT-LBP-DCT, the DWT-LBP features and the LBP-DCT features for face-spoofing, respectively. The LBP-DCT features have been applied in our previous work [7].
Conclusion and future work
The paper is aimed to address the issue of face anti-spoofing by using an advanced feature DWT-LBP-DCT and a machine learning classifier SVM. The DWT with a segmentation operation is applied to decompose images into different frequency components, the LBP operation is utilized to extract the spatial information, and the DCT operation is performed to take advantage of energy concentration property to obtain the temporal information by efficiently combining multiple video frames. Finally, the
CRediT authorship contribution statement
Wanling Zhang: Conceptualization, Methodology, Software, Data curation, Writing - original draft. Shijun Xiang: Conceptualization, Methodology, Writing - original draft, Validation.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (No. 61772234) and the Special Funds for the Cultivation of Guangdong College Students’ Scientific and Technological Innovation, China (No. pdjh2020a0060).
Wanling Zhang was born in JiangXi, China, in 1997. She received the B.S. degree from Jinan University, China, in 2018. She is currently pursuing the Master degree in Jinan University, China. Her current research interests include face anti-spoofing and reversible data hiding.
References (32)
- et al.
Integration of image quality and motion cues for face anti-spoofing
J. Vis. Commun. Image Represent.
(2016) - et al.
3D Visual passcode: Speech-driven 3D facial dynamics for behaviometrics
Signal Process.
(2019) - et al.
Wavelet leaders and bootstrap for multifractal analysis of images
Signal Process.
(2009) - et al.
How far did we get in face spoofing detection?
Eng. Appl. Artif. Intell.
(2017) - et al.
Personal identification based on iris texture analysis
IEEE Trans. Pattern Anal. Mach. Intell.
(2003) - et al.
Evaluation of speaker verification security and detection of HMM-based synthetic speech
IEEE Trans. Audio Speech Lang. Process.
(2012) - et al.
A survey on antispoofing schemes for fingerprint recognition systems
Acm Comput. Surv.
(2015) - et al.
Online signature verification based on Kolmogorov-Smirnov distribution distance
- et al.
Detection of video-based face spoofing using LBP and multiscale DCT
- et al.
Face spoofing detection from single images using micro-texture analysis
On the effectiveness of local binary patterns in face anti-spoofing
Face spoofing detection from single images using texture and local shape analysis
Iet Biometr.
Cited by (24)
Face spoofing detection using Heterogeneous Auto-Similarities of Characteristics
2024, Engineering Applications of Artificial IntelligenceFace spoofing detection based on multi-scale color inversion dual-stream convolutional neural network
2023, Expert Systems with ApplicationsDeep malware detection framework for IoT-based smart agriculture
2022, Computers and Electrical EngineeringFace morphing attack detection and attacker identification based on a watchlist
2022, Signal Processing: Image CommunicationCitation Excerpt :In many countries, facial Automated Border Control (ABC) systems are widely deployed at airports and seaports to improve the efficiency of border crossings [1,2].
A Review of Face Detection Anti Spoofing Techniques on Varied Data Sets
2024, Lecture Notes in Electrical Engineering
Wanling Zhang was born in JiangXi, China, in 1997. She received the B.S. degree from Jinan University, China, in 2018. She is currently pursuing the Master degree in Jinan University, China. Her current research interests include face anti-spoofing and reversible data hiding.
Shijun Xiang received the B.S. degree, the M.S. degree and the Ph.D. degree from Changan University (1997), Guizhou University (2000), and Sun Yatsen University (2006) of China, respectively. From 2006 to 2007, he was a Post-Doctoral Researcher with Korea University, Seoul, South Korea. He is currently a Full Professor with the School of Information Science and Technology, Jinan University, Guangzhou, China. He has authored or co-authored over 60 peer-reviewed papers, including the IEEE TCSVT and the IEEE TMM. His current research interests include robust watermarking, reversible data hiding, secure signal processing in encrypted domain, and face spoofing.