Skip to content

Quality Metric Algorithms for Fingerprint Images

Latent print examiners (LPEs) recognize the connection between “accuracy of call” and both the number of features and their “quality.” In fact, in the very first step of the ACE-V process, “assessment,” the LPE provides a subjective assessment of the adequacy of the latent print for forensic identification purposes. At present, this “A” step is based on the LPE who judges the “clarity” or “quality” of the features in the print for going forward with the remaining C, E, and V steps of the analysis. The ultimate goal of the ACE-V process is an accurate determination of the evidence, either in identifying the perpetrator of the crime or in excluding a possible person of interest.

SWGFAST developed a “sufficiency chart” that showed contours for “poor”, “adequate”, “very good” accuracy as a function of number of minutiae in a latent fingerprint image and their “quality.” The contours in the SWGFAST chart were based on “expert opinion” rather than on an objective set of numerical scores that characterize quality, and, ultimately, provide guidance on the accuracy of the final assessment if the print is subjected to the full ACE-V analysis. An objectively defined quality metric, and a data-based relationship between the value of the metric and the accuracy of the final assessment by and examiner (following the other 3 steps in the ACE-V process) is essential, both for (a) confidence in the examiner’s final assessment; and (b) lab process operations. (With limited LPE resources, a lab manager may ask LPEs spend time on only high-quality prints, that lead to acceptably high probabilities of assessment accuracies, rather than spend time on low-quality prints.)

To assist Lab Mangers in the “A” part of the latent print analysis process, this webpage provides several available quality metric algorithms. The Contrast algorithm (see https://sctyner.github.io/OpenForSciR/fingerprints.html), DFIQI (Swofford), and wavelet-Based SNoQE (Huckemann-Richter-Thai) algorithms are fully open-source. Links are provided to access the other algorithms. Additional links provide information and details for all algorithms. New algorithms will be posted as they are developed.

A CSAFE project is underway to associate print quality (as quantified through scores on these quality metrics) with “probability of accurate assessment” via a double-blind study. The results of that project will be reported when the study is complete. The study results will have immediate practical value, both in providing a confidence coefficient in the examiner’s final assessment, and in assisting lab managers with lab process operations (e.g., low-quality prints can be given low priority without other evidence in the case).

Biometric Special Databases and Software

Available from the National Institute of Standards and Technology (NIST) at www.nist.gov/itl/iad/image-group/resources/biometric-special-databases-and-software.

Friction Ridge Databases

Special Database 302: Nail to Nail (N2N) Fingerprint Challenge (user’s guide available).

Special Database 301: Nail to Nail (N2N) Fingerprint Challenge Dry Run (user’s guide available).

Special Database 300: Uncompressed Plain and Rolled Images from Fingerprint Cards (user’s guide available).

Contrast Gradient Algorithm

fingerprint r package (Contrast Gradient Algorithm)

To Install from Github

library(devtools)

install_github(“kdp4be/fingerprintr”)

Simple Example Using Data (image + minutiae information) Included in the Package

G080_image <- read.bmp(“data/G080_image.bmp”)
# convert to format algorithm expects
image_file <- convert_image(G080_image, “bmp”)
# load minutiae, already in matrix format
data(G080_min)
# quality scores
quality_scores(image_file, G080_min)

Relevant Papers

A New Measurement for the Quality of Individual Minutiae in Latent Fingerprints

Open Forensic Science in R (September 19, 2019, also available at https://github.com/sctyner/OpenForSciR)  

LQMetric

Requests for the Universal Latent Workstation (ULW) software should be submitted through FBI/CJIS. You can request the software at www.fbibiospecs.cjis.gov/Latent/DownloadSoftware.

NOBLIS believes that FBI/CJIS will distribute ULW to U.S. agencies freely, and to researchers on a case-by-case basis.

To Install

Follow install instructions.pdf for either of the ULWSoftware*.zip files. (Note: ULWSoftwarePlusGround.zip gives an additional ground truth folder, which should be unzipped and in this ULW Files folder as well.)

Running

LQMetricsUserGuideWithCmdLine.pdf contains run instructions, command line section is in Part G at the end. From the installed folder, run LQMetric.exe image.ext. Adding -v -h is recommended to obtain column names and more features, e.g., LQMetric.exe Sample_004_G.BMP -v -h

Commands such as the following can be used to loop over a folder of images and create CSV files of minutiae information:

  • for /f %f in (‘dir /b .\500\’) do LQMetric.exe -v .\500\%f >> output500.txt
  • for /f %f in (‘dir /b .\nist_sd27a_prints\’) do LQMetric.exe .\nist_sd27a_prints\%f -LM=.\nist_sd27a_minutiae\%f.csv
  • for /f %f in (‘dir /b .\GroundTruth\Images\Latents1000ppi\’) do LQMetric.exe .\GroundTruth\Images\Latents1000ppi\%f -LM=.\groundtruth_minutiae\%f.cs

Relevant Papers

Changes in Latent Print Fingerprints Examiners’ Markup between Analysis and Comparison (2014)

Assessing the Clarity of Friction Ridge Impressions (2013)

Understanding the Sufficiency of Information for Fingerprint Value Determinations (2013)

DFIQI

DFIQI Software (written by Henry Swofford, January 29, 2021)

Relevant Reference

NIST Data Annotated DFIQI LQMetric
This spreadsheet shows a sample DFIQI output for raw NIST SD27 images. The first few columns are the “raw” output measurements of the five variables. The next set of columns labeled “Norm-” is a normalized score function of the raw output variables. The LQSraw is simply the mean of the normalized scores. The LQMPixColorClassificaton column is how LQMetric color coded the region where the minutiae occurred.

Additional Resources

Quality Metrics for Pattern Evidence Poster (Karen Pan and Karen Kafadar, 2019)