Face Recognition

Model by Paravision

This model detects all faces in a given image and creates an embedding—a biometric representation—for each detected face. Embeddings can be added to a database (i.e. enrollment) or matched against a database of registered embeddings to perform verification or identification functions. The key functions provided include:

  • Face detection: Detecting the presence and location of all faces in a submitted image. At this stage, key locations on the face (“Landmarks”) will be noted, and a quality score will be calculated to determine if the image of a face under consideration is sufficiently high quality to ensure a high level of face recognition matching accuracy.
  • Embedding generation: Computation of a set of unique mathematical vectors (“Embeddings”) for every face.
  • Face enrollment: Adding an embedding to a database when a face is presented for registration.
  • Face matching: Comparing the embedding for a given face against an enrollment database.
  • Description

    Product Description


    Paravision face recognition has been extensively benchmarked by the National Institute of Standards and Technology (NIST) Face Recognition Vendor Tests (FRVT), which are the gold-standard global benchmarks for face recognition testing. These tests include very large, statistically significant data sets testing a variety of conditions, from Visa (travel document) quality to fully unconstrained. Paravision has performed exceedingly well across all tests, including 1:1 Verification, 1:N Identification, and Image Quality Metrics. Paravision is ranked #1 in the U.S., UK, and Europe on all major NIST FRVT leaderboards, including 1:1, 1:N, and face mask effects. Paravision has demonstrated superior accuracy in the most challenging benchmarks, delivering #3 global performance 1:N identification, #2 global performance in face mask effects, and #1 global performance in profile (90 degree) matching.


    In order to deliver the demonstrated level of accuracy, Paravision has developed a highly sophisticated machine learning infrastructure, combining very large and diverse datasets with the latest generation AI technologies. In addition to training, Paravision has also developed a highly robust set of internal benchmarks that assess performance across imaging conditions as well as subject demographics such as age, gender, and ethnicity. By pairing best-of-breed ML and biometric performance benchmarking, Paravision has shown an ability to consistently perform at a world-class level and continuously improve even over short release intervals.