Counter-measures to photo attacks in face recognition: A public database and a baseline
read more
Citations
Distillation as a Defense to Adversarial Perturbations Against Deep Neural Networks
Face Spoof Detection With Image Distortion Analysis
On the effectiveness of local binary patterns in face anti-spoofing
A face antispoofing database with diverse attacks
Distillation as a Defense to Adversarial Perturbations against Deep Neural Networks
References
The Nature of Statistical Learning Theory
Pattern Recognition and Machine Learning (Information Science and Statistics)
A direct adaptive method for faster backpropagation learning: the RPROP algorithm
Face Recognition with Local Binary Patterns
The DET Curve in Assessment of Detection Task Performance
Related Papers (5)
Frequently Asked Questions (12)
Q2. What is the method for detecting spoofing?
Motion-based algorithms for anti-spoofing typically use complex methods such as Optical Flow estimators to extract deformation patterns from the image being analyzed.
Q3. What is the missing key to this puzzle?
A missing key to this puzzle is the lack of standard databases to test counter-measures, followed by a set of protocols to evaluate performance and allow for objective comparison.
Q4. What are examples of detectable texture patterns?
Examples of detectable texture patterns are printing failures or overall image blur. [12] describes a method for print-attack detection by exploiting differences in the 2-D Fourier spectra comparing the hard-copies of client faces and realaccesses.
Q5. How many quantities are used to describe the signal pattern for windows of N nonoverlapping images?
To input the motion coefficients into a classifier and avoid the variability in time, the authors extract 5 quantities that describe the signal pattern for windows of N nonoverlapping images.
Q6. How do the authors combine the time information with the binary decision scheme?
In order to combine the time information with that of the window-based classifier, the authors accumulate the output over time for every block of N frames and apply a very simple binary decision scheme using a majoritywins approach.
Q7. What is the likely source of the input data coming from a spoof attempt?
If there is no movement (fixed support attack) or too much movement (hand-based attack), the input data is likely to come from a spoof attempt.
Q8. How many images can be used to calculate the MD?
MD = 1SD ∑ (x,y)∈D |It(D)− It−1(D)| (2)The calculation of MD, even considering both RoIs, can be implemented in a very efficient manner allowing the variable to be computed for every two images in the sequence being observed.
Q9. What is the common variable in research?
One variable often disregarded in research is the motion pattern introduced by the attacker, while displaying the device with the photograph of the client face being attacked.
Q10. What is the definition of a spoofing detection system?
A spoofing detection system is subject to two types of errors, either the real access is rejected (false rejection) or an attack is accepted (false acceptance).
Q11. What is the first work on literature to propose a publicly available database specifically tailored towards the development?
This is the first work on literature to propose a publicly available database specifically tailored towards the development of spoofing counter-measures. [13] present a technique to evaluate liveness based on a short sequence of images using a binary detector that evaluates the trajectories of selected parts of the face presented to the input sensor using a simplified optical flow analysis followed by an heuristic classifier.
Q12. How does the HTER on the test set compare to the values on the fixed-support column?
The authors again confirm their expectations that the system would work better for hand-based attacks by observing that the HTER on that subset is always smaller or equal to the values on the fixed-support column.