VISTeam (ISR UC) achieved the top result in the NIST Face Recognition Vendor Morph Test (FRVT MORPH - the world's biggest benchmark for estimating the performance of automated face morph detection
VISTeam solution demonstrated a great performance in a target no-reference (by a single image) face morphing detection achieving the top result in a number of benchmark datasets by the evaluating metrics APCER@BPCER=(0.1, 0.01) (Attack Presentation Classification Error Rate at a certain level of Bona Fide Classification Error Rate). Namely it outperforms competitors in following tests: “Low Quality Morphs - Website”, “Automated Morphs - Local Morph”, “Automated Morphs - Local Morph Colorized Match”, “Automated Morphs - Visa-Border”, and also in the challenging one “High Quality Morphs - Manual”.
Also the VISTeam morphing detection approach gives the most regular result across various benchmark datasets and doesn't demonstrate bias to any particular morphing method.
Additionally, the VISTeam solution achieved the top result for the APCER@BPCER=0.01 in differential morphing detection even with zero effort of adapting the solution to the differential scenario.
The FRVT MORPH test provides ongoing independent evaluation, which is designed to obtain a commonly measured assessment of morph detection capability to inform developers and end-users. The test leverages a number of datasets created using different morphing methods with goals to evaluate algorithm performance over a large spectrum of morphing techniques. Testing was conducted using a tiered approach, where algorithms were evaluated on low quality morphs created with readily accessible tools available to non-experts, morphs generated using automated morphing methods based on academic research, and high quality morphs created using commercial grade tools.