Skip to content

Latest commit

 

History

History
207 lines (169 loc) · 12.9 KB

README_FINGER.md

File metadata and controls

207 lines (169 loc) · 12.9 KB

PADISI-Finger Dataset

This document analyzes the PADISI-Finger dataset introduced in Multi-Modal Fingerprint Presentation Attack Detection: Evaluation On A New Dataset and provides
instructions on how to download and use the dataset. The provided dataset is only the PADISI-USC portion of the dataset. We are not authorized to release the PADISI-APL portion of the dataset, as mentioned in the manuscript.

Dataset organization

Ground truth file

The ground truth file is located under data/finger_partitions/padisi_USC_FINGER_ground_truth.csv. The .csv file contains the following columns:

  • transaction_id: Alphanumeric code defining a single data collection session for a participant.

  • trial_id: Alphanumeric code defining each captured finger of the participant.

  • trial_name: One of ['FINGER_RIGHT_THUMB', 'FINGER_RIGHT_INDEX', 'FINGER_RIGHT_MIDDLE', 'FINGER_RIGHT_RING', 'FINGER_LEFT_THUMB', 'FINGER_LEFT_INDEX', 'FINGER_LEFT_MIDDLE', 'FINGER_LEFT_RING'] specifying which finger was captured.

  • modality: Biometric modality (FINGER).

  • ground_truth: Ground truth code defined as:

    • m00000: Bona-fide
    • m4<PAI_CODE><PAI_SPECIES>: Presentation attack with PAI_CODE = [01, 02, ..., 29], according to Table 1 below. PAI_SPECIES = [01, ..., MAX_PAI_SPECIES_FOR_CODE], according to Table 1 below (e.g., [01, ..., 04] for PAI_CODE = 01). Example: m40502 represents a presentation attack of the 2nd species of code 05, which is a Gummy material finger. Note that only the USC columns of Table 1 are relevant for the provided PADISI-USC portion of the dataset.

  • participant_id: Alphanumeric code uniquely defining each participant.

  • gender, ethnicity, race, age: Demographic information about the participant for each sample (see statistics in Table 2).

Partition files

All partition files are also located under data/finger_partitions with descriptive names. They involve all 3FOLD and LOO protocols presented in the paper. Each .csv file contains the following columns:

  • transaction_id, trial_id, trial_name: Same as in the ground truth, above. Can be used to uniquely associate each row of a partition .csv file to a row in the ground truth .csv file.
  • partition_name: One of ['train', 'valid', 'test'] specifying if a sample belongs to the training, validation or testing set, respectively.
Table 1 PAI counts in the collected datasets. For each PAI code, we provide a general PAI description, the number of different species as well as the attributes used for grouping PAI codes in terms of material, species, transparency, and attack type. PAI categories whose appearance depends heavily on the participant and preparation method are marked with *. Sponsor approval is required to release additional information about each PAI code.

Table 2 PADISI-USC dataset statistics and demographics.

Downloading the dataset

  1. Please download and sign the provided Data Transfer and Use Agreement. Both the recipient (must be a project's principal investigator) and an official authorized signatory of the recipient’s organization must sign the agreement. For example, at a university, an authorized signatory is typically an administrative official, rather than a student or faculty member. PLEASE, CAREFULLY READ THIS PAGE.
  2. Submit the request and upload your signed Data Transfer and Use Agreement at PADISI-USC Finger Dataset Request.
  3. You will receive the download link and instructions upon approval and you can download the dataset within 30 days from approval.

If you have any questions about the data request process you can send an email to:

using [PADISI USC Finger]: Dataset request question on the subject line.

Using the dataset (No custom package installation required)

For ease of use, we are providing a preprocessed version of our PADISI-USC dataset (used in the experiments presented in Multi-Modal Fingerprint Presentation Attack Detection: Evaluation On A New Dataset). When you download the preprocessed data, you will receive a file padisi_USC_FINGER_preprocessed.bz2 which can be loaded in Python using the joblib package (see conda installation or pip installation) as:

import joblib
data_dict = joblib.load("padisi_USC_FINGER_preprocessed.bz2")

The above command will load a python dictionary with the following entries:

"data"       : 4d numpy array of shape (6211, 24, 160, 80)  
               - 6211: total number of samples
               -   24: total number of image channels per sample - each channel is normalized in 0-1 based on camera's bit depth (see paper for details).
               -  160: image height
               -   80: image width
"identifiers": list of 6211 string identifiers for all samples (unique string identifier per sample)
"labels"     : list of 6211 boolean labels for all samples
"pai_codes"  : list of 6211 string ground truth codes for all samples
  • "data": Contains all 24 channels of FM, FS, FL, and BN data, as presented in the paper:
    • FL: Channels 0-9 (10 frames of LSCI data - frames 10-19 out of 100 total frames).
    • FS: Channels 10-13 (4 images of SWIR data with order [1200nm, 1300nm, 1450nm, 1550nm]).
    • FM: Channels 14-20 (3 images of Visible and 4 of NIR data with order [465nm, 591nm, white, 720nm, 780nm, 870nm, 940nm]).
    • BN: Channels 21-23 (3 frames of Back-Illumination data - frames 10-12 out of 20 total frames).
  • "identifiers": Each identifier is constructed as ID_<transaction_id>_<trial_id>_<trial_name> using the entries of the ground truth or partition files above. This identifier can be used to uniquely associate each sample in the "data" with their corresponding ground truth or partition.
  • "labels": False: Bona-fide, True: Presentation attack.
  • "pai_codes": Ground truth code, using the rules discussed above.

Example data loading (Requires installing custom packages)

We are providing a simple example for looping through the preprocessed data using custom packages. The example also shows the creation of a PyTorch DataLoader. This requires creating an environment using conda and installing the provided package. The example code also works without PyTorch (if not installed).

  • Installation instructions:

    1. Use Anaconda3 or Miniconda3

    2. Create the padisi environment:

      conda env create -f padisi-environment.yml

      or

      conda env create -f padisi-environment-with-pytorch.yml (if you want to use PyTorch).

    3. Activate the environment and install the package:

      conda activate padisi
python setup.py install

  • Running the example (see script for flag explanation and additional available flags):

    conda activate padisi
python finger_data_example.py -dbp ../../data/finger_partitions/padisi_USC_FINGER_dataset_partition_3folds_part0.csv -extractor_id FM_FS

Licence

The dataset and code is made available for academic or non-commercial purposes only.

USC MAKES NO EXPRESS OR IMPLIED WARRANTIES, EITHER IN FACT OR BY OPERATION OF LAW, BY STATUTE OR OTHERWISE, AND USC SPECIFICALLY AND EXPRESSLY DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, VALIDITY OF THE SOFTWARE OR ANY OTHER INTELLECTUAL PROPERTY RIGHTS OR NON-INFRINGEMENT OF THE INTELLECTUAL PROPERTY OR OTHER RIGHTS OF ANY THIRD PARTY. SOFTWARE IS MADE AVAILABLE AS-IS. LIMITATION OF LIABILITY. TO THE MAXIMUM EXTENT PERMITTED BY LAW, IN NO EVENT WILL USC BE LIABLE TO ANY USER OF THIS CODE FOR ANY INCIDENTAL, CONSEQUENTIAL, EXEMPLARY OR PUNITIVE DAMAGES OF ANY KIND, LOST GOODWILL, LOST PROFITS, LOST BUSINESS AND/OR ANY INDIRECT ECONOMIC DAMAGES WHATSOEVER, REGARDLESS OF WHETHER SUCH DAMAGES ARISE FROM CLAIMS BASED UPON CONTRACT, NEGLIGENCE, TORT (INCLUDING STRICT LIABILITY OR OTHER LEGAL THEORY), A BREACH OF ANY WARRANTY OR TERM OF THIS AGREEMENT, AND REGARDLESS OF WHETHER USC WAS ADVISED OR HAD REASON TO KNOW OF THE POSSIBILITY OF INCURRING SUCH DAMAGES IN ADVANCE.

Bibliography

If you use this dataset, please cite the following publications:

[1] Leonidas Spinoulas, Hengameh Mirzaalian, Mohamed Hussein, and Wael AbdAlmageed, “Multi-Modal Fingerprint Presentation Attack Detection: Evaluation On A New Dataset”, in IEEE Transactions on Biometrics, Behavior, and Identity Science, vol. 3, no. 3, pp. 347-364, July 2021, doi: 10.1109/TBIOM.2021.3072325

[2] Leonidas Spinoulas, Mohamed Hussein, David Geissbühler, Joe Mathai, Oswin G. Almeida, Guillaume Clivaz, Sébastien Marcel, and Wael AbdAlmageed, “Multispectral Biometrics System Framework: Application to Presentation Attack Detection”, in IEEE Sensors Journal, vol. 21, no. 13, pp. 15022-15041, July 2021, doi: 10.1109/JSEN.2021.3074406

Bibtex format

@article{Spinoulas2021a,
  author    = {Spinoulas, Leonidas and
               Mirzaalian, Hengameh and
               Hussein, Mohamed E. and
               AbdAlmageed, Wael},
  title     = {{Multi-Modal Fingerprint Presentation Attack Detection: Evaluation On A New Dataset}},
  journal   = {IEEE Transactions on Biometrics, Behavior, and Identity Science},
  year      = {2021},
  volume    = {3},
  number    = {3},
  pages     = {347-364},
  doi       = {10.1109/TBIOM.2021.3072325}
}

@article{Spinoulas2020b,
  author    = {Spinoulas, Leonidas and
               Hussein, Mohamed E. and
               Geissb{\"u}hler, David and
               Mathai, Joe and
               Almeida, Oswin G.
               and Clivaz, Guillaume
               and Marcel, S{\'e}bastien
               and AbdAlmageed, Wael},
  title     = {{Multispectral Biometrics System Framework: Application to Presentation Attack Detection}},
  journal   = {IEEE Sensors Journal},
  year      = {2021},
  volume    = {21},
  number    = {13},
  pages     = {15022-15041},
  doi       = {10.1109/JSEN.2021.3074406}
}

Acknowledgment

This research is based upon work supported by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via IARPA R&D Contract No. 2017-17020200005. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the ODNI, IARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon.