Rationale
Software-based detection of biometric presentation attacks is also called liveness detection, or anti-spoofing. In previous years, investigations on the PAD (presentation attack detection) and matching systems' embedding problem considered parallel and sequential combinations of PAD and matcher. However, performance degradation was highlighted after such embedding and, in order to assess if such integrated system is still effective for security applications, it would be necessary to evaluate how much is this degradation. While the sensor characteristic and material adopted are the main variables to evaluate a PAD system, the spoofing attack probability should be considered as well. For example, in consumer applications, such as sensors embedded into smartphones, vendors could hypothesize that presentation attacks are much less likely or not relevant than in high-security applications. Therefore, in order to investigate such embedding problem, it would be necessary to use a tool modeling the presentation attack probability would allow evaluating, for example, the impact of a specific type of attack (e.g., using a specific material for creating fake samples) and the sensor technology on the whole system’s ROC. Thanks to this tool, we would be able to decide for which operational points and conditions the given embedding is worthy to be implemented or not.
To this aim, Micheletto et al. proposed a simulator based on the probabilistic modeling of relationships among variables at hand in the case of the sequential fusion of presentation attacks detector and matcher. This simulator takes as input the ROC curves of the matcher and the PAD. The output is the whole acceptance rate in its three essential components: the genuine users one (GAR), the zero-effort attacks one (FMR), and the presentation attacks one (IAPMR). Two parameters are added: the probability of a presentation attack (w in our webapp, which denotes the relative cost of presentation attacks with respect to zero-effort impostors) and the PAD’s operational point (APCER ≤ p% or BPCER ≤ p%). Using this simulator does not require implementing or replicate any PAD algorithm or matching system. This is also what a designer would prefer to do: take the vendors’ individual ROCs and explore the performance achievable according to some expected scenarios. Therefore, since a PAD is tailored over the specific sensor, the simulator helps the designer select the most appropriate technology for the final application’s security level.
According to the proposed model, the GFMR (Global False Match Rate) of the integrated system is:
Where is a boolean event which define if the access is granted to a certain user when the matching score
between the input image and the user’s claimed identity templates is over a given acceptance threshold
, while
is another boolean event which defines if the liveness detector gives the classification of a certain input sample as alive when the liveness score
, obtained by the analysis of the feature set extracted from the input image, is over a certain liveness threshold
Therefore, we included this simulator into a webapp, which allows users to use it without the need to reproduce such model, in an intuitive and straightforward way.
Go to the other sections of the Wiki for further information about the simulator.