Release notes

These release notes are relevant for the following products:

• MegaMatcher 13.x Standard / Extended SDK
• VeriFinger 13.x Standard / Extended SDK
• VeriLook 13.x Standard / Extended SDK
• VeriEye 13.x Standard / Extended SDK
• VeriSpeak 13.x Standard / Extended SDK

You may subscribe for newsletters to receive notifications about the released updates.

List of releases:

• March 16, 2023
• March 13, 2023

See also the list of release notes for previous versions.

Product revision number: 20230316
Product revision hash: e4fab3ad62c4a9a5112eb3b7003d1eff1b801fd5

Fixes:

• Fixed several minor bugs and issues.

Product revision number: 20230313
Product revision hash: e4fab3ad62c4a9a5112eb3b7003d1eff1b801fd5

The first release of MegaMatcher 13.0 SDK line.

Algorithms-related changes:

• Fingerprint recognition:
  • Fingerprint quality estimation with new features added such as fingerprint wetness/dryness/pressed to hard/pressed too soft.
  • Support optional matching of fingers with different scales, such as of child vs adult.
  • Support optional matching of mirrored fingerprints and returns appropriate status (if mirrored matched) on matching details.
  • Matching details also returned matched finger position.
• Facial recognition:
  • Faces template creation and matching algorithms have been significantly improved. In particular, faces with masks handling is a lot better. While templates created with 13.0 line give the best accuracy, older product line 12.X and 11.X face templates are still supported and can be matched in any combination. Face templates from product line 10.X and earlier are no longer supported.
  • Face image quality is now calculated more accurately and allows to prevent accidental extraction of non-face or low quality images. Lots of face additional attributes are calculated when processing face image. Some of them are useful for more detailed quality evaluation (overexposure, compression artifact detection, etc.), while others enable application specific logic (smile, glasses, head covering detection, etc.). Most of the attributes are now in easy to understand 0..100 range (0 - no/not present, 100 - yes/present). Deprecated lots of complicated legacy attribute checking patterns which required multiple field checking in order to get correct results.
  • ICAO and face liveness check algorithms optimized for practical use. Not only face liveness algorithm is compliant with ISO 19794-5 presentation attack detection standard, but actually passing liveness check with lots of different devices and various environmental settings it is now easier for user.
• Iris recognition:
  • Irises extractor and matcher were enhanced with additional biometric features which improve recognition accuracy in difficult conditions. In particular, when only a small part of the eye is visible. While 13.0 line iris templates are matched more accurately, compatibility with previous product lines is preserved.
  • Iris validation allow more accurate checking if provided image indeed contains iris. Validation can be performed both on images and existing templates.
  • One possible problem with iris scanning includes large rotation angle (most often 180 degrees when scanner is held upside-down). Algorithm can determine angle of rotation automatically (NEAttributes::RotationAngle). If NBiometricEngine parameter "Irises.UseAngle" is turned on, then iris is also automatically rotated back. Note: in case iris scanner is held upside-down, left and right iris positions are swapped. Therefore it is needed to update position in such cases: either switch to opposite position, or make it unknown.
• Speaker recognition:
  • Algorithm is now working on both 8 kHz and 16 kHz recordings without explicit parameter.
  • Significantly improved speed for matching millions of voiceprints per second on MegaMatcher Accelerator.

New features:

• Experimental GPU acceleration support – the template extraction algorithms for all modalities received experimental support for GPU acceleration for NVidia PC cards and Jetson ARM boards. TensorRT 8.5.2 dependency needs to be installed by user.
• Signature scanners' support have been integrated into NDeviceManager and some device module plugins for specific device integration added: Wacom, Mip, SignoTec. Captured signatures are returned in forms of vector (if possible) or image.

Updates and improvements:

• Removed backwards compatibility with 10.X line face templates.
• Removed 32-bit platform support for Microsoft Windows and Linux (x86).
• Removed x86 Intel Android support.