NIST testing shows Rank One facial recognition algorithms’ combination of speed and accuracy
Rank One Computing has announced that its biometric algorithms have reduced error rates in NIST’s Ongoing Face Recognition Vendor Test (FRVT) by 10 to 50 percent on all datasets while improving match speed.
The company points out that in a scatter plot of accuracy vs. algorithm speeds in the NIST report, Rank One’s accuracy is shown to be within a percentage point of the top algorithms while performing ten times faster. In the most recent testing, Rank One’s “rankone-005” algorithm, which corresponds with version 1.17 of the ROC SDK, had mugshot accuracy higher than the top algorithm in April testing.
The combination of accuracy and efficiency is critical for developing systems that process streaming images or run on mobile devices, according to the announcement. Efficiency also enables law enforcement and government customers quickly re-enroll large databases.
Rank One provides its technology on a perpetual licensing model, with a small maintenance fee giving customers access to all algorithm updates.
NIST says that advances in convolutional neural networking have pushed the accuracy of facial recognition algorithms to 20 times better accuracy than only 5 years ago.
Rank One’s facial recognition technology was recently integrated into a turn-key multimodal biometric system from Radiant Solutions.