August 25, 2015 -
CyberExtruder announced it has released its new Aureus 3D facial reconstruction software, as well as version 5.0 of its facial recognition matching algorithm.
The company claims that Aureus 3D facial reconstruction software successfully improves on facial recognition matching accuracy by 26.5% when compared to some of the leading 2D facial recognition products.
The facial recognition matching accuracy results is taken from tests for the U.S. Department of Homeland Security, which illustrates how Aureus 3D’s ability to generate a 3D representation from 2D images greatly improves the technology for suspect and victim identification.
“Unique solutions like this become an essential resource for the intelligence, security and law enforcement communities,” said Jack Ives, co-founder and chief operating officer for CyberExtruder. “Aureus 3D is expected to be a valuable new tool for those engaged in mission critical facial recognition.”
Aureus 3D captures the subject’s facial image and reconstructs it into a 3D format that takes into account the many poses and facial expressions, as well as compensates for low lighting and partial images.
In reconstructing a face, Aureus 3D’s patented processes are able to determine the person’s original pose relative to the camera.
The software uses this data to generate results which lead to matches when a subject’s face is turned away from the camera by as much as 70 degrees and a downward (or upward) angle of 25 degrees.
The solution’s resulting 3D model, which is called a 3D facial template, is used to perform identification and verification tasks.
“Aureus 3D analyzes one (or more) recorded images of a person’s face and uses its vast knowledge of the human face, with its understanding of environmental factors, to craft a very unique 3D model of that person,” said Ives. “It’s a claim no other software package can make.”
Aureus 3D software, which is compatible with any facial recognition system, can either be added to existing surveillance networks or used as a standalone workstation.
CyberExtruder’s Aureus 3D version 5.0 is the latest update to its facial recognition matching algorithm, which significantly improves facial recognition by combining Aureus 3D’s patented 2D/3D pose correction technology with deep convolutional neural networks.
The new 3D product enables OEMs and system integrators to benefit from greater facial recognition performance, allowing them to deliver the industry’s highest-performing face matching solutions.
The Aureus 3D product release is uniquely suited for facial identification in real time video surveillance applications, low resolution images and video, bookmarking and retrieval of facial sequences archival video, comprehensive understanding of faces in crowds, and wearable face matching solutions.
CyberExtruder’s new algorithm delivers significant performance increases over algorithms previously documented in NIST’s 2013 MBE testing, including a 15.8% improvement in matching accuracy over v4.0, 17.7% better matching accuracy than COTS matcher on LFW, 22.1% better matching accuracy on LFW than Facebook’s Deep Face, matching speed of 10 million matches per second, and a template size of 256 bytes.
Aureus 3D version 5.0 is able to accommodate non-front facing images and real-time video, faces that are partially occluded, and low-resolution video.
Additionally, the 3D product delivers great results in applications where both the subject and camera are in motion such as body worn cameras.