Facial Cosmetics Database and Impact Analysis on Face Recognition

Get Complete Project Material File(s) Now! »

Spoofing in Face Recognition

In a spoofing attempt, a person tries to masquerade as another person and thereby, tries to gain access to a recognition system. Since face data can be acquired easily in a contactless manner, spoofing is a real threat for face recognition systems.
There are several types of spoofing attacks such as photograph, video or mask attacks. Due to its 3D face shape characteristics, the detection of 3D mask attacks appears to be more challenging compared to the detection of 2D attacks such as photograph and video attacks. In this section, spoofing in face recognition is analyzed in two groups as 2D spoofing attacks and countermeasures and 3D spoofing attacks and countermeasures.

2D Spoofing Attacks and Countermeasures

The most common spoofing attacks to face recognition systems are achieved by using photographs and videos due to their convenience and low cost. It has been shown that face recognition systems are vulnerable to photograph and video attacks.
For instance, in the study of Määtta et al. [87], it is stated that the Windows XP and Vista laptops of Lenovo, Asus and Toshiba can be spoofed easily. These laptops authenticate users by scanning their faces. According to [87], at Black Hat 2009 conference, the Security and Vulnerability Research Team of the University of Hanoi demonstrated how to easily spoof the systems « Lenovo’s Veriface III », « Asus’ SmartLogon V1.0.0005 », and « Toshiba’s Face Recognition 2.0.2.32 » using fake facial images of the user. In [87], it is also claimed that this vulnerability is now listed in the National Vulnerability Database of the National Institute of Standards and Technology (NIST) in the US. This simple example demonstrates that countermeasure techniques have to be developed to enhance the security and robustness of face biometric systems. Figure 2.1 shows an example for photograph spoofing.

3D Spoofing Attacks and Countermeasures

Compared to 2D spoofing attacks such as photograph and video, 3D mask attacks to face recognition systems is a considerably new subject. The main reason for the delay in mask spoofing studies is due to the unavailability of public mask databases.
The preparation of a mask attacks database is much more difficult and expensive than the preparation of photograph or video attacks databases. Initially, to prepare a high quality mask, a 3D scanner is necessary to obtain the 3D model of the target person, which are generally high-cost devices. The procedure continues with manu facturing of the masks which is also an expensive procedure.
When 3D masks are introduced as attacks, some of the countermeasures proposed for the detection of 2D spoofing attacks are no more applicable. The study of Kollreider et al. [71] shows that a face recognition system relying on eye blinking and lip movements can be defeated by using photographic masks wrapped over face with eyes and mouth regions cut out. Also, since motion based countermeasures depend on different movements of 2D and 3D surfaces, they are not applicable when masks are used instead of photos or videos. It appears that the detection of 3D mask attacks is more challenging compared to the detection of 2D facial attacks.
If a 3D mask is not able to spoof a recognition system, it is not a successful attack, and there is no need to develop a countermeasure against it. Therefore, initially the spoofing performances of the masks have to be analyzed. Unfortunately, there are only two studies in the literature which evaluates the spoofing performances of 3D mask attacks. In our study [74], we analyzed how well the spoofing performances of the masks used in our studies are. To the best of our knowledge, the spoofing performances of mask attacks on both 2D and 3D face recognition were first analyzed in our study [74] using the mask database which was prepared within the context of the European Union (EU) research project TABULA RASA [111]. The mask database used in our study [74] includes both 2D and 3D data hence we were able to evaluate the impact of mask attacks on both 2D and 3D face recognition. Since this database consists of very high-quality masks, they have almost perfect spoofing performances especially in 3D. After our study which examines spoofing performances of 3D mask attacks [74], very recently a new study is published by Erdogmus et al. [41], which presents the 3D Mask Attack Database (3DMAD). 3DMAD was recorded with a low-cost depth camera (Microsoft Kinect sensor). In [41], they also analyzed the spoofing performances of 3D mask attacks using 3DMAD database however only for 2D face recognition, which is the main limitation of the study. The reported results in [41] demonstrate that facial masks can pose a serious threat to 2D face recognition systems.

READ  The stem/progenitor cell niche in the adult human hypothalamus

Table of contents :

Abstract
Acknowledgments
Contents
List of Figures
List of Tables
List of Publications
1 Introduction 
1.1 Motivation
1.2 Original Contributions
1.3 Outline
2 Face Spoofing and Disguise Variations 
2.1 Introduction
2.1.1 Biometrics
2.1.2 Face Recognition
2.2 Spoofing in Face Recognition
2.2.1 2D Spoofing Attacks and Countermeasures
2.2.2 3D Spoofing Attacks and Countermeasures
2.3 Disguise Variations in Face Recognition
2.3.1 Facial Alterations and Countermeasures
2.3.2 Facial Makeup and Countermeasures
2.3.3 Facial Accessories (Occlusions) and Countermeasures
2.4 Conclusions
3 Countermeasure Technique Against Photograph Attacks 
3.1 Introduction
3.2 The Photograph Database
3.3 The Proposed Approach
3.3.1 Pre-Processing
3.3.2 Feature Extraction
3.4 Experiments and Analysis
3.4.1 Test 1: Results under Illumination Change
3.4.2 Test 2: Effect of DoG Filtering in the Proposed Approach
3.5 Conclusion
4 Impact of Mask Spoofing on Face Recognition 
4.1 Introduction
4.2 The Mask Database
4.3 The Selected Face Recognition Systems
4.3.1 Pre-Processing
4.3.2 Face Recognition Systems
4.4 Experiments and Analysis
4.5 Conclusions
5 Countermeasure Techniques Against Mask Attacks 
5.1 Introduction
5.2 Countermeasure Techniques Against Mask Attacks
5.2.1 Techniques used Inside the Proposed Countermeasures
5.2.2 The Proposed Countermeasures
5.3 Experiments and Analysis
5.3.1 Stand-Alone Classification Performances of the Countermeasures
5.3.2 Integration of the Countermeasure to 3D Face Recognition System
5.4 Conclusions
6 Impact of Nose Alterations on Face Recognition 
6.1 Introduction
6.2 Simulating Nose Alterations
6.3 Experiments and Analysis
6.3.1 Impact on 2D Face Recognition
6.3.2 Impact on 3D Face Recognition
6.4 Conclusion
7 Face Recognition Robust to Nose Alterations 
7.1 Introduction
7.2 Block Based Face Recognition Approach Robust to Nose Alterations
7.3 Experiments and Analysis
7.3.1 Evaluation on 2D Face Recognition
7.3.2 Evaluation on 3D Face Recognition
7.4 Conclusion
8 Facial Cosmetics Database and Impact Analysis on Face Recognition
8.1 Introduction
8.2 Facial Cosmetics Database
8.2.1 Specification of the Database
8.2.2 Acquisition of Images
8.2.3 Structure of the Database
8.3 Facial Cosmetics
8.3.1 Specification of Facial Cosmetics
8.3.2 Classification of Applied Facial Cosmetics
8.4 Evaluation and Results
8.4.1 Test Setup
8.4.2 Evaluation Results
8.5 Conclusion
9 Kinect Face Database and Impact Analysis on Face Recognition 
9.1 Introduction
9.2 Review of 3D Face Databases
9.3 The Kinect Face Database
9.3.1 Database Structure
9.3.2 Acquisition Environment
9.3.3 Acquisition Process
9.3.4 Post-Processing
9.3.5 Potential database usages in addition to face recognition
9.4 Benchmark Evaluations
9.4.1 Baseline Techniques and Settings
9.4.2 Pre-processing
9.4.3 Evaluation Protocol
9.4.4 Evaluation Results
9.4.5 Fusion of RGB and Depth Face Data
9.5 Data Quality Assessment of KinectFaceDB and FRGC
9.6 Conclusion
10 Conclusions and Future Perspectives 
10.1 Summary
10.2 Future Work
Résumé Etendu en Français
11 Résumé Etendu en Français 
Bibliography

GET THE COMPLETE PROJECT

Related Posts