High-Precision Face Photo Quality Checks
Face SDK 3.28 introduces two new quality estimation modules:
Quality Control – Reduces recognition errors by automatically filtering out low-quality photos (blurred, too small, rotated, etc.) before they enter the pipeline. The upgraded core modification evaluates parameters such as face size, frontalness, noise and others for both recognition and Liveness detection, returning a clear pass/fail verdict with a detailed list of failed checks. This feature is especially valuable for remote identification and access control systems.
Quality Assessment – Ensures compliance with international standards, including ISO/IEC 19794-5:2011 and ICAO Guidelines, by applying a comprehensive set of quality checks.
3X Faster Face Database Search
With updated DynamicTemplateIndex and MATCHER_MODULE, Face SDK 3.28 makes face database searches up to three times faster. In addition, new support for AVX-512 instructions provides an extra performance boost on compatible processors.
+14% More Accurate Liveness Detection
The fifth version of the LIVENESS_ESTIMATOR (2d_ensemble) increases attack detection accuracy by approximately 14% at default thresholds. Under stricter target BPCER levels (reduced from 5% to 1%), the new module outperforms all previous Liveness modules, providing stronger protection against spoofing attempts.
Ready for Real-World Applications
Whether preventing fraud in digital onboarding or powering ultra-responsive access control systems, Face SDK 3.28 is designed to deliver fast, scalable, and deepfake-resistant facial recognition from day one.
More information about 3DiVi Face SDK: https://3divi.ai/products/software/face-sdk
The full documentation: https://docs.3divi.ai/face_sdk/overview About 3DiVi Inc.
3DiVi Inc., founded in 2011, is one of the leading international developers of AI and machine learning (ML) technologies for computer vision. The company’s computer vision algorithms cover face recognition, body and skeletal tracking, spatial understanding and object recognition.
