How to evaluate accuracy of fingerprint recognition algorithm?

#1 Data collection and processing

Fingerprint recognition is closely related to data, which access to its protects. The secret and reliability of the security method lies in the quality and diversity of the dataset used. It should cover a wide range of fingerprint variations, including different individuals, ages, and environmental conditions. To ensure realistic algorithm testing, the presence of normalisation, enhancement, and segmentation is required.

#2 Performance metrics

The accuracy of fingerprint recognition is controlled by a few performance metrics. It’s enough to start with the False Acceptance Rate (FAR), which measures the probability of granting access to an unauthorised user. The opposite situation is checked by the False Rejection Rate (FRR), when a legitimate user may not get access to the system. There is also a possibility of checking the rate of correctly accepted legitimate users, called the Genuine Acceptance Rate (GAR).

In situations where the FAR and FRR are equal, the Equal Error Rate (EER) is used to ensure a balanced measure of performance. Another metric worth mentioning is the Receiver Operating Characteristic (ROC) Curve, which represents the trade-off between FAR and FRR across different threshold values in a graphical way.

#3 PFT evaluations by NIST

The National Institute of Standards and Technology (NIST) conducts a series of assessments on the Proprietary Fingerprint Template (PFT) technology, focusing on one-to-one fingerprint matching. These evaluations are designed to measure the accuracy and performance of the PFT algorithm using proprietary templates, unified datasets, and standardized rulesets. To facilitate testing, NIST introduced PFT III, enabling participants to assess the algorithm’s reliability by utilising biometric data supplied by NIST. Moreover, NIST plays a crucial role in preparing the testing environment and protocol for these evaluations.

#4 Cross-validation

To ensure the highest quality of the fingerprint recognition reliability evaluation, it is quite common to use cross-validation. The process required the division of the dataset into multiple subsets, training the algorithm on a subset, and testing it on the remaining subsets. The process is repeated multiple times for a more reliable assessment of the algorithm’s performance. You can read more about fingerprint recognition algorithm evaluation to delve deeper into the topic.