FVC-onGoing
SourceAFIS algorithm has been submitted for evaluation to FVC-onGoing competition. Latest results are shown in the table below.
| Benchmark area | Benchmark | Algorithm | Version | EER | FMR100 | FMR1K | FMR10K | Zero FMR | Details |
|---|---|---|---|---|---|---|---|---|---|
| Fingerprint Verification | FV-STD | SourceAFIS | 3.14.0 | 3.87 % | 4.73 % | 6.55 % | 9.05 % | 10.6 % | submission 8426 |
| Fingerprint Verification | FV-HARD | SourceAFIS | 3.14.0-net | 8.19 % | 11.6 % | 15.1 % | 19.4 % | 21 % | submission 8427 |
Analysis
The results show that SourceAFIS is on the lower end of the accuracy spectrum, but they also show that SourceAFIS can perform very well with high-quality fingerprints.
Notice that FNMR at zero FMR is around 10%. That is most likely because SourceAFIS has trouble matching the lowest quality 5% of fingerprints, but it can match the remaining 95% with high confidence.
Applications that need to keep FNMR low should either ensure high fingerprint quality during enrollment or capture multiple fingers per subject.
Sources
FVC-onGoing requires participants to implement command-line protocol. Source code for wrappers implementing this protocol on top of SourceAFIS is opensourced in repositories listed below.
| Benchmark area | Executable | Repository | Engine |
|---|---|---|---|
| Fingerprint Verification | enroll.exe |
sourceafis-fvc-fv-extractor | SourceAFIS for .NET |
| Fingerprint Verification | match.exe |
sourceafis-fvc-fv-matcher | SourceAFIS for .NET |