DESCRIPTION OF DeepSignDB
The DeepSignDB database comprises a total of 1526 users from four different popular
databases and a novel signature database not presented yet, named e-BioSign DS2. Fig. 1 graphically
summarises the design, acquisition devices, and writing tools considered in the DeepSignDB database.
Figure 1. Description of the design, acquisition devices, and writing tools considered
in the new DeepSignDB database. A total of 1526 users and 8 different captured devices are used (5
Wacom and 3 Samsung general purpose devices). For the Samsung devices, signatures are also collected
using the finger. Gen. Sig. = Genuine Signatures, and Sk. Forg. = Skilled Forgeries.
A short description of each database regarding the device, writing input, number of acquisition
sessions and time gap between them, and type of impostors is included inside the paper
[ArXiv2020_DeepSign].
STANDARD EXPERIMENTAL PROTOCOL
The DeepSignDB database has been divided into two different datasets,
one for the development and training of the system and the other one for the final evaluation.
The development dataset comprises around 70% of the users of each database whereas the remaining
30% are included in the evaluation dataset. It is important to note that each dataset
comprises different users in order to avoid biased results. Thus, we first identified all those
users that took part in the acquisition of different databases.
For the training of the systems, the development dataset comprises a total of
1084 users. In our experiments carried out in [ArXiv2020_DeepSign], we have divided this dataset
into two different subsets, training (80%) and validation (20%). However, as this dataset is used
only for development, and not for the final evaluation of the systems, we prefer not to set any
restriction and let researchers use it as they like.
For the final testing of the systems , the remaining 442 users of the
DeepSignDB database are included in the evaluation dataset. In order to perform a complete
and fair analysis of the signature verification systems, and see their generalization capacity
to different scenarios, aspects such as the inter-session variability, number of training
signatures, impostor scenario, and writing input (stylus/finger) have been considered in the final
experimental protocol design.
Table 1 describes all the experimental protocol details of the DeepSignDB
evaluation dataset for both stylus (top) and finger (bottom) writing inputs.
Table 1: Experimental protocol details of the DeepSignDB evaluation dataset (442
users). Numbers are per user and device.
All data provided is organised as follows:
DeepSignDB
|-- Development
|-- Finger
|-- Stylus
|-- Evaluation
|-- Finger
|-- Stylus
Comparison_Files
|-- Finger
|-- 1vs1
|-- Random
|-- Skilled
|-- 4vs1
|-- Random
|-- Skilled
|-- Stylus
|-- 1vs1
|-- Random
|-- Skilled
|-- 4vs1
|-- Random
|-- Skilled
We provide all details regarding comparison files and nomenclature in the "README.pdf" file send it
when applying for the DeepSignDB database.
EVALUATION RESULTS
Table 2 and 3 provide the evaluation performance results of our proposed TA-RNN
approach for the whole DeepSignDB evaluation dataset and for each of the databases
included in it when using the stylus and finger as input. In addition, we compare the proposed
TA-RNNs [TCYB2019_DeepSign] with the preliminary benchmark results presented in [ArXiv2020_DeepSign]
with other popular approaches such as DTW and RNNs for completeness. All details are included in
[ArXiv2020_DeepSign].
Table 2: System performance results (EER) over the DeepSignDB evaluation dataset.
Stylus scenario.
Table 3: System performance results (EER) over the DeepSignDB evaluation dataset.
Finger scenario.
REFERENCES
For further information on the database and on different applications where it has been
used, we refer the reader to (all these articles are publicly
available in the publications section
of the BiDA group webpage.)
-
[TBIOM2021_DeepSign] R. Tolosana, R. Vera-Rodriguez, J. Fierrez, and J. Ortega-Garcia, "DeepSign: Deep On-Line Signature Verification", IEEE Transactions on Biometrics, Behavior, and Identity Science, 2021.
-
[PAA2010_BiosecurID] J. Fierrez, J. Galbally, et al., "BiosecurID: A Multimodal Biometric
Database", Pattern Analysis and Applications, Vol. 13, n. 2, pp. 235-246, May 2010.
-
[TPAMI2010_BiosecureDS2] J. Ortega-Garcia, J. Fierrez, et al., "The Multi-Scenario
Multi-Environment BioSecure Multimodal Database (BMDB)," IEEE Trans. on Pattern Analysis and
Machine Intelligence, vol. 32, no. 6, pp. 1097- 1111, 2010.
-
[VISP2003_MCYT] J. Ortega-Garcia, J. Fierrez, et al., "MCYT Baseline Corpus: A Bimodal Biometric
Database", IEEE Proc. Vision, Image and Signal Processing, Vol. 150, n. 6, pp. 395-401,
December 2003.
Please remember to reference articles [ArXiv2020_DeepSign, PAA2010_BiosecurID,
TPAMI2010_BiosecureDS2, VISP2003_MCYT] on any work made public, whatever the form, based directly or
indirectly on any part of the DeepSignDB database.
