研究目的
To address the ear presentation attack detection problem for the first time by developing a benchmarking study and proposing a new ear artefact database using light field imaging.
研究成果
The paper successfully introduces the first ear PAD database and benchmarking study, showing that light field based solutions achieve high accuracy and low computational complexity. Future work should expand the database to include 3D PAIs for more comprehensive assessment.
研究不足
The database only includes artefact samples from digital display PAIs (2D attacks), not 3D PAIs like wrapped paper or silicon ears, which may limit generalization to all attack types. The extended set has a smaller sample size, potentially affecting performance.
1：Experimental Design and Method Selection:
The study involves benchmarking state-of-the-art light field and non-light field based PAD solutions using a newly created ear artefact database. Methods include texture-based, quality-based, focus/depth-based, and learning-based approaches, with SVM classifiers and cross-validation protocols.
2：Sample Selection and Data Sources:
The Lenslet Light Field Ear Artefact Database (LLFEADB) is used, consisting of baseline and extended sets with bona fide and artefact samples captured from 67 and 15 subjects respectively, using a Lytro ILLUM camera. Artefacts are created using PAIs like laptop, tablet, and mobile phones.
3：List of Experimental Equipment and Materials:
Lytro ILLUM lenslet light field camera, MacBook Pro 13'', iPad Air2 9.7'', iPhone 6S, Sony Xperia z2, MATLAB R2015b, Lytro Desktop Software, Matlab Light Field Toolbox v.0.
4：7'', iPhone 6S, Sony Xperia z2, MATLAB R2015b, Lytro Desktop Software, Matlab Light Field Toolbox v.Experimental Procedures and Operational Workflow:
4. 4. Experimental Procedures and Operational Workflow: Artefact acquisition involves displaying 2D bona fide images on PAIs and capturing with the Lytro camera. Performance evaluation uses 4-fold cross-validation with training on 3/4 of data and testing on 1/4, repeated 50 times for average results.
5：Data Analysis Methods:
Performance metrics include BPCER, APCER, and ACER. Computational complexity is assessed in terms of feature extraction and classification times per image, using MATLAB on a standard PC.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容-
Lytro ILLUM
ILLUM
Lytro
Used for capturing light field images of ears and artefacts in the database.
-
MacBook Pro
13 inch
Apple
Used as a presentation attack instrument to display 2D bona fide images.
-
iPad Air2
9.7 inch
Apple
Used as a presentation attack instrument to display 2D bona fide images.
-
iPhone 6S
6S
Apple
Used as a presentation attack instrument to display 2D bona fide images.
-
Sony Xperia z2
z2
Sony
Used as a presentation attack instrument to display 2D bona fide images.
-
Lytro Desktop Software
4
Lytro
Used for rendering 2D central view images from light field raw files.
-
Matlab Light Field Toolbox
v.0.4
MathWorks
Used for processing light field images, rendering 2D images, and creating multi-view arrays.
-
MATLAB
R2015b
MathWorks
Used for implementing PAD solutions, data analysis, and time measurements.
-
登录查看剩余6件设备及参数对照表
查看全部
