Neurotechnology was started with the key idea of using neural networks for various applications such as biometric person identification, computer vision, robotics, and artificial intelligence. Much to our delight, we were able to endure the “neural networks winter” by using and expanding this expertise all through 2012, the year that brought explosive developments in the concept and infrastructure of deep neural networks. In turn, this allowed us to quickly take advantage of the emerging opportunities that came with the new wave of deep learning. This approach to computing has triggered an entire range of new projects in object recognition and other applications. Currently, our team comprises 100+ employees, 15% of them holding a Ph.D., and half of the employees being involved in R&D activities.
Neurotechnology was founded in Vilnius, Lithuania in 1990. Next year we released our first fingerprint identification system for criminal investigations. Our further research endeavor resulted in the first fingerprint identification algorithm for civil uses that was made public in 1997. Also, our researchers got involved in developing a solution for recognizing faces – starting in 2002, and releasing the first product in 2004. This was followed by our algorithm for iris recognition released in 2008. In addition, we have an on-going research program on voice recognition since 2011.
Once we conceived the benefits of fusing several biometric modalities, we directed our efforts towards building a multi-biometric product. It was released in 2005 under the name MegaMatcher Software Development Kit. The initial version could support recognition of fingerprints and faces; the modalities for iris and voice recognition were added later. Right from the date of its release in 2005, our MegaMatcher SDK gained acceptance as a key solution in large and national-scale projects related to issuance of passports, de-duplication of voters, etc. For instance, our customers in Bangladesh used the technology for de-duplication of the voters in the 2008 election. As another example, the MegaMatcher SDK has been used for passport issuance of the Indonesian nationals since 2009. There are many more similar projects as well.
Since applications of this type put extreme requirements on the 1:N matching performance, we have spent a lot of our R&D effort on solving this problem. To speed the matching process, we released MegaMatcher Accelerator in 2009. Originally it was used for the fingerprint modality only, but then we added the face and iris modalities. In 2016, we released MegaMatcher ABIS to provide the best solution for customers willing to have end-user software instead of an SDK. In 2018, we used MegaMatcher ABIS during the large-scale voter de-duplication campaign in the Democratic Republic of the Congo. We carried out a number of similar projects in other countries, too.
In response to the market demands, since 2000 we have also been involved in developing a range of products for smartcard-based biometrics and embedded applications, as well as some end-user products.
In 2004, to better accommodate the growing volumes of research in artificial intelligence, the company founded its robotics division that began research in the field of mobile autonomous robots. In 2009, to assist our customers in selecting suitable hardware for their implementations of biometric systems, we started Biometric Supply. This subsidiary offers biometric scanners of 140+ models from multiple manufacturers that are all supported by the software from Neurotechnology.
The year 2012 was rich of events for Neurotechnology. To take advantage of the new opportunities brought by rapid growth of cloud technologies, we started SkyBiometry. This subsidiary now provides face detection and recognition software as a cloud-based service. Also in 2012, a strategic decision was made to start a division in Sri Lanka. The team of developers there is now the prime force in developing our biometric solutions for attendance systems.
In 2014, Neurotechnology released SentiBotics – a ready-to-use robotics development kit. The same year, the company established the Ultrasound Research Group to undertake research in the fields of ultrasonic particle manipulation, parametric array, and transducer technology.
Neural networks and AI
In almost 30 years of our activity we have accumulated substantial experience in the area of neural networks that allowed us to develop a multitude of functionalities for our products based on deep learning as well as those for our customer applications.
Most frequently, deep neural networks are able to solve many problems – for instance, image classification, object detection, or instance segmentation – more efficiently than traditional computer vision algorithms. To facilitate this process, we have built a technology that has tools to support developing AI-based object recognition applications.
In the area of face recognition, the last few years witnessed a dramatic reduction in error rates brought by new algorithms that are based on convolutional neural networks (CNN). Our company started using CNNs for the task of face recognition in 2013. The first application of neural networks alone resulted in an improvement of the accuracy of unconstrained face recognition by a factor of 15 times! We expect further improvements in face recognition performance due to explosive development of architectures and techniques related to CNNs.
As deep learning techniques kept proliferating other areas, we were ultimately able to employ our face recognition algorithms under the conditions of real-time surveillance. By being able to recognize and track other objects – such as pedestrians and all kinds of vehicles (cyclists, bikes, cars, buses, trucks, etc.) - in adjacent video frames, we can extract various information about those objects (for example, the color of the vehicle, or the direction of its movement). There is a separate modality for recognizing license plates of vehicles using neural networks.
Our latest improvements in other biometric modalities are also driven by extensive research in deep neural networks. This includes the most interoperable fingerprint algorithm in the world (ranked the first in the NIST MINEX interoperability category), the second most accurate iris recognition technology, and the new version of our algorithm for speaker identification.
As another illustration of our achievements in using deep learning, Neurotechnology’s researchers won the first place in the 2017 Kaggle competition with a computer vision solution for classifying fish species.
We seek to distill our knowledge and understanding of how the natural intelligence operates into deep learning-based algorithms, and we see this approach as the shortest path towards achieving the General AI. Neurotechnology also runs Deep Learning Paper Reviews, a series of open-access events devoted to sharing and discussing recent ideas in the field.
Biometric technologies and products
We have a reputation for developing various products for biometric identification of fingerprints, palm prints, face, irises, and voice. Since the release of our first fingerprint identification system, we have delivered 200+ products and version upgrades for identification and verification of objects and personal identity. More than 3,000 system integrators, security companies and hardware providers in 140+ countries integrate our algorithms into their products.
With a combination of fast algorithms and high reliability, company's fingerprint, face, iris and voice biometric technologies can be used for access control, computer security, banking, time attendance control and law enforcement applications, among others.
With millions of customer installations worldwide, our products are used for both civil and forensic applications, including border crossings, criminal investigations, systems for voter registration, verification and duplication checking, passport issuance and other national-scale projects. Neurotechnology's fingerprint identification algorithms have shown one of the best results for reliability in major biometric competitions and evaluations, including the National Institute of Standards & Technology (NIST) Minutiae Interoperability Exchange III (MINEX III), Proprietary Fingerprint Template Evaluation II (PFT II) and Fingerprint Vendor Technology Evaluation for the US Department of Justice (FpVTE 2012). Previously, the fingerprint recognition algorithms have received awards in the International Fingerprint Verification Competitions (FVC2006, FVC2004, FVC2002 and FVC2000).
In 2018, our iris recognition algorithm has been tested in the NIST Iris Exchange (IREX) Evaluation and recognized as the second most accurate among those tested. The accelerated version of the algorithm was nearly 50 times faster than any other matching system in the NIST IREX IX evaluation. Previously, Neurotechnology showed outstanding results in the IREX, IREX III and IREX IV evaluations.
To meet the demands of a variety of applications, we developed many advanced algorithms based on computer vision. For instance, they are used in SentiSight.ai, our interactive web platform for developing AI-based object recognition applications. The platform has tools designed to support interactive model training without coding and faster image labeling, thus reducing the amount of user effort spent on data mining. We also run projects tailored to the specific needs of our customers. Our technology for real-time surveillance was designed to support biometric face identification of moving pedestrians using live video streams from high-resolution digital cameras. The technology is used for passive identification – when passers-by do not make any efforts to be recognized. List of possible uses includes law enforcement, security, attendance control, visitor counting, traffic monitoring, and other commercial applications.
Some of our ealier R&D efforts related to computer vision also involved technologies for eye movement tracking and 3D object model reconstruction.
Our current R&D effort in robotics is focused towards the “programming by demonstration” approach and its practical implementation in various neural network structures. During the experiments we not only gained experience with different types of neural networks (e.g., various CNN’s, RNN’s, including RNN’s with external memory, the self modifying RNN, and the CMP, Cognitive Mapping and Planning), but also developed our own theoretical innovations.
SentiBotics, our ready-to-use robotics development kit, allows rapid development and testing of mobile robots. The kit includes a mobile robotic platform with a 3D vision system, a modular robotic arm and accompanying ROS-based software we have built on our own, with complete source code and programming samples.
Ultrasound research group
Ultrasound Research Group undertakes research in the fields of ultrasonic particle manipulation, parametric array and transducer technology. It develops novel algorithms, hardware and electronics solutions for ultrasonics applications. Currently, the group is developing a new, patent-pending 3D printing and assembling technology based on ultrasonic particle manipulation. This new technology is intended to expand capabilities of existing 3D printing and assembling processes. In 2018, the group also filed a patent for a novel ultrasonic electrostatic transducer technology. These transducers were developed primarily for use in parametric array systems.