PFT II and PFT III Evaluations
Neurotechnology has constantly participated in the NIST PFT evaluations. Different versions of the fingerprint recognition algorithms were submitted and showed top results there. The latest submissions to the PFT II and the ongoing PFT III are in average the most accurate algorithms in all the experiments
The Proprietary Fingerprint Template (PFT) is a series of fingerprint matching technology evaluations, which are run by NIST to assess one-to-one matching performance and accuracy with proprietary templates while using unified datasets and rulesets. The most recent PFT III is designed in a way that the testing is performed by each participant with NIST-provided biometric data, using NIST-provided testing environment and protocol. The results are then submitted to NIST for further evaluation. See PFT III official page for more information.
The current PFT III includes datasets from the previous PFT II, so the comments are provided for both evaluations to show the evolution of Neurotechnology's fingerprint recognition algorithm:
NIST PFT II
These comments provided by Neurotechnology are based on NIST PFT II results, published on November 19, 2018.
NIST Proprietary Fingerprint Template Evaluation II (PFT II) was one-to-one verification evaluation which measures the performance of fingerprint matching algorithms by utilizing proprietary fingerprint templates. The samples dataset was increased to 120,000 subjects compared to previous PFT evaluation. Number of experiments was also increased to 33 with different combinations of single and two fingerprints matching.
In 2018 Neurotechnology fingerprint algorithm was submitted to NIST Proprietary Fingerprint Template Evaluation II. The algorithm 's template matching accuracy was among the best participants in most of the experiments.
Our latest submission to the PFT II was 4E. In summary, we present the accuracy results as FNMR at FMR = 0.0001 for each of the 33 experiments:
- 1-9 correspond to plain-to-plain fingerprints matching on AZLA dataset.
- 10-18 correspond to plain-to-rolled fingerprints matching on AZLA dataset.
- 19-27 correspond to rolled-to-rolled fingerprints matching on AZLA dataset.
- 28-30 correspond to plain-to-plain fingerprints matching on DHS2 dataset.
- 31-33 correspond to plain-to-plain fingerprints matching on POEBVA dataset.
We also present template size, template extraction and comparison times for five most accurate submissions for all four different datasets:
Our latest submission 4E (in cyan) was among the most accurate algorithms in all experiments. It was also the second fastest during enrollment and second/third fastest during matching of fingerprint templates.
NIST PFT III
These comments provided by Neurotechnology are based on NIST PFT II and PFT III results, reviewed on April 19, 2022.
The PFT II evaluation ended in 2019 and has been changed with PFT III which includes more databases and experiments. But it also includes comparison with the previous PFT II experiments.
We present the accuracy results as FNMR at FMR = 0.0001 for each of the 33 experiments:
- 1-9 correspond to plain-to-plain fingerprints matching on AZLA dataset.
- 10-18 correspond to plain-to-rolled fingerprints matching on AZLA dataset.
- 19-27 correspond to rolled-to-rolled fingerprints matching on AZLA dataset.
- 28-30 correspond to plain-to-plain fingerprints matching on DHS2 dataset.
- 31-33 correspond to plain-to-plain fingerprints matching on POEBVA dataset.
Our latest submission Neurotechnology+0003 (in cyan) is in average the most accurate algorithm in all the experiments at the PFT III. Also we can see evolution of our algorithm since 2015 when we started participating in the PFT II and continued to the PFT III. Our accuracy has been constantly increasing in almost all of the tests.
