Figure 1.
Example of a signature acquisition for the Signature Long-Term DB using the Wacom Intuos 3 digitizing tablet and a paper template with a delimited signing area for each sample.
Figure 2.
General time diagram of the different acquisition sessions that conform the Signature Long-Term Database.
Figure 3.
Typical samples that can be found in the Signature Long-Term DB.
Each signature corresponds to each of the acquisition sessions of five different users.
Figure 4.
Performance of the three signature recognition systems used in the experiments, considering 4 signatures for enrollment, and evaluated on the BiosecurID DB.
Table 1.
Division of the feature set introduced in [56] (given also in Appendix S1) according to the type of information they contain.
Table 2.
Enrollment and test signatures used to compute the genuine scores in the aging experiments.
Figure 5.
Performance evolution of the three signature recognition systems considered in the experiments.
For the DTW-based system only two curves appear as for experiments A–C its EER is close to zero. The EER for the three systems and for the different experiments are reported in Table 3.
Table 3.
EER for the aging experiments defined in Table 2. The whole DET curves for these experiments are shown in Fig. 5.
Figure 6.
Evolution through time of the mean (circles) and variance (vertical lines) of the genuine score distributions (in vertical on the right) for the three systems considered in experiments A–E.
A darker gray level represents a better performance of the given system.
Figure 7.
Aging Coefficient (AC) from the least affected to the most affected user by aging in the Signature Long-Term DB, for the three systems considered in the experiments.
Please note that the least affected user, the most affected user, or any of the users in between, do not necessarily have to coincide (i.e., be the same signer) for all three systems. The three AC curves are shown on the same figure for an easier visual comparison across systems.
Table 4.
Most and least affected users by aging in the Signature Long-Term DB according to the three systems considered in the experiments.
Figure 8.
Most (left) and least (right) affected users by aging in the Signature Long-Term DB according to Table 4.
Table 5.
Enrollment and test signatures used to compute the genuine scores in the template update experiments.
Figure 9.
Mean (circles) and variance (vertical lines) of the genuine score distributions (in vertical on the right) for the 4 different template update strategies tested and for the three systems considered in the experiments.
A darker gray shade represents a better performance of the given system.
Figure 10.
Evolution through time of the duration, maxima points in x, maxima points in y, number of penups and speed of the signatures in the Signature Long-Term Database.
Figure 11.
Variation Coefficient (VC) from the least variable to the most variable dynamic and static features (see Table 1) proposed in [56] (see Appendix S1).
Table 6.
Most and least variable features over time.