NIST SD302: Data Exploration and Filtering

NIST Special Database 302 has been split into several parts from SD302a to SD302i. Our focus is on NIST SD302b and NIST SD302h exclusively.

The NIST SD302b and NIST SD302h databases do not contain completely matching fingerprint pairs. Therefore, we filtered the datasets to identify and include only the matching fingerprint pairs between both databases.

We provide a CSV containing the list of filtered files from NIST SD302 at the following link: NIST_SD302h_mate.csv. Details of the data filtering process are outlined below.

This repository is part of the project available at SFP-Progressive-Feedback-Latent-Fingerprint-Restoration.

Data Exploration

NIST SD302b

The SD 302b directory structure is organized as follows:

images
    - baseline                   # Collection type. Contain device code R, S, U, and V.
        - R                      # Device code
            - 1000               # Resolution
                - slap           # Impression type
                    - png        # Image format
                - slap-segmented
                    - ...
            - 500
                - ...
        - ...

Count the subjects for each device.

We explore the subject in the database by counting the subject or person by uisng count_subject.py

Image names are in the form SUBJECT_ACTIVITY_HAND_ENCOUNTER_TECHNIQUE_DIGITIZER_RESOLUTION_DEPTH_CHANNELS_LPNUMBER_SOURCE.EXT

Device Code	Number of subjects
R	92
S	108
R and S	200
U	200
V	200

NIST SD302h

The SD 302h directory structure is organized as follows:

ebts                        # Record format
    - latent                # Collection type
        - lffs              # Transaction type
            - original     
                - 1000      # Resolution, in PPI
            - enhanced
                - ...
    - checksum_latent_lffs_enhanced.csv # File checksums
    - checksum_latent_lffs_original.csv

Data Filtering

In data_filter.ipynb, We perform filtering on latent fingerprint images from NIST SD302h using finger position labels from finger_positions.csv, considering corresponding mates present in NIST SD302b (devices: U and V)

Convert LFFS files to Images

The LFFS format includes Field Number 13.999, indicating a LATENT FRICTION RIDGE IMAGE. We read LFFS files starting from 13.999 until IEND flag. These contents are passed into an io.BytesIO object created from the encoded image's bytes. We utilize Image.open from the PIL library to open and save the image into files. The source code for this process is available in lffs_to_image.py.

Refrence

• Data Format for the Interchange of Fingerprint, Facial & Other Biometric Information ANSI/NIST-ITL 1-2011 NIST Special Publication 500-290 Edition 3

• Special Publication (NIST SP) - 500-290e3

  • [NIST.SP.500-290e3.pdf] (https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.500-290e3.pdf)

• SYNTHETIC LATENT FINGERPRINT GENERATOR

Citing SFP

If you are using SFP or benchmarks in your research, kindly reference DOI: 10.1109/ACCESS.2024.3397729 the following.

@ARTICLE{10526230,
  author={Kriangkhajorn, Supakit and Horapong, Kittipol and Areekul, Vutipong},
  journal={IEEE Access}, 
  title={Spectral Filter Predictor for Progressive Latent Fingerprint Restoration}, 
  year={2024},
  volume={12},
  number={},
  pages={66773-66800},
  keywords={Fingerprint recognition;Image restoration;Friction;Frequency-domain analysis;Filtering;Image matching;Deep learning;Image restoration;Image forensics;Machine learning;Fingerprint recognition;image restoration;image enhancement;image filtering;image forensics;machine learning},
  doi={10.1109/ACCESS.2024.3397729}}

or

S. Kriangkhajorn, K. Horapong and V. Areekul, "Spectral Filter Predictor for Progressive Latent Fingerprint Restoration," in IEEE Access, vol. 12, pp. 66773-66800, 2024, doi: 10.1109/ACCESS.2024.3397729.

Contact

If you have any questions or need assistance, reach us at supakit.kr@gmail.com.