Company

Smart Engines – Smart AI in everything

Company was formed in 2010 by a group of image processing and recognition specialists with a goal of making complex document analysis and recognition technologies available for mobile platforms. Development from the very beginning was aimed to on-device recognition, and not to traditional “cloud recognition”. For the customer this leads to several advantages: since there is no large file transfer the response speed is higher (and, if Internet traffic is non-free, the cost is lower). The main advantage is higher information security – a trust between the customer and the bank doesn’t necessarily lead to a trust between the customer and an unknown third-party cloud service provider used by the bank.

In order to achieve higher on-device recognition precision and speed Smart Engines specialists had to reject most traditional approaches used in OCR systems and concentrate on merging of hierarchical synthesis architecture for the fuzzy model of recognized object “as a whole”, technologies for ultra-large-scale training datasets synthesis for neural network models, and engineering approaches of algorithmic and arithmetic computations optimization of neural networks and image processing algorithms.

Even the detection and location of the document boundaries on the video frame in real time using a low-end smartphone CPU is a technologically-difficult problem. A solution for this problem lead to the release in 2011 of the first company’s product – PDF-scanner for smartphones. Later, in 2015, the document tracking algorithm used in this system came third in ICDAR2015 Competition on Smartphone Document Capture and OCR (SmartDoc) – a competition held by organizers of the main scientific conference on document recognition.

Based on previous experience in 2014 the company released a system for machine-readable zones recognition which was capable of identity checking using any passport or ID which contains MRZ. In this project two technologies developed by Smart Engines were first used – integer neural networks with small bit count and optimal inter-frame integration of recognition results in video stream. This gave the opportunity to use the product even on the cheapest smartphones, which is an important factor for deployment in large companies and government structures.

Both embossed bank cards and documents containing MRZ have high degree of standardization which greatly reduced the complexity of development and testing of the recognition systems. A creation of a product which is capable of any document recognition is a task on another complexity level. Since the number of broadly used document types is measured in hundreds in order to develop such a system in realistic timeframes, technologies of one-shot learning are required. These technologies have to be applied at least in two key modules – document identification and geometrical location and text fragment recognition.

The latter is not obvious because it is generally accepted that the task of text recognition is solved many years ago. However in 2015 Smart Engines released the pioneer system for recognition of all fields of Russian Internal Passport for mobile phone which does not use any cloud OCR technologies, and the recognition quality was found to be 2.5 higher compared to cloud OCR product of a widely known company which develops text recognition systems for many years and holds a dominant position on this market. In 2016 Smart Engines released a bank cards recognition product which supports not only embossed cards, but also cards manufactured with indent-printing technology. The recognition quality in this case also significantly exceeded generic OCR systems. In order to achieve these results a technology is developed for completely synthetic training datasets creation based on a small number of real samples. This technology is still actively used in all projects of Smart Engines.

The solution for the former problem – detection, identification, location and determining orientation of documents in video stream based on a single sample – allowed Smart Engines to demonstrate at Mobile World Congress 2017 a universal system of on-device ID documents recognition in real time. Now an addition of new document type support to the product takes only several days.

In the process of development and evolution of our products and technologies specialists of Smart Engines created a complete set of image processing and recognition libraries optimized for a broad range of computational platforms. Not a single line of third-party code is used in modules pertained to the company mission. This approach, albeit highly resource-consuming, in the long run allows better fine-tuning of all used algorithms for any specific project and continuous growth of developers’ qualification.

Company management consciously stimulates academic and scientific activity of employees. In the last 3 years company’s specialists obtained 5 patents and published more than 130 scientific papers. Members of the board of directors participate in international conferences as experts and organizers.

Currently the company has 45 employees, among them students and aspirants of leading Russian universities, 14 PhDs and Doctors of Science. Senior employees teach in Moscow Institute of Physics and Technology on a specialized department. CSO of Smart Engines is a corresponding member of Russian Academy of Sciences. Interesting fact about Smart Engines is that it employs two of the four authors of widely known “Method of Four Russians” (published in 1970).

Management

Arlazarov Vladimir V.
Chief Executive Officer

Ph.D., Leading specialist in development and application of recognition systems and data mining. Works in IT-related fields since 1997. Managed more than 70 projects related to creation and embedding of document recognition systems for Russian Pension Fund, FMS and EMERCOM of Russia, SBERBANK, Moscow Metro, Gazprom structures and others companies. Under his leadership several generations of industrial OCR and document capture systems were developed, including the first industrial system for Russian internal passport recognition. Leader of a team of developers who created a system for personalization of new generation passports and visa documents. Author of several scientific papers and patents.
vva@smartengines.biz

Nikolaev Dmitry P.
Chief Technology Officer

Ph.D., Associate Professor. Finished an internship in SAIT (South Korea). Leading Russian scientist in image processing, machine vision and pattern recognition fields. Author of more than 150 scientific publications. Chief architect of cross-platform package of high-performance image processing and recognition libraries “MIN*”. Leading developer of the first industrial system for Russian internal passport recognition. According to nVidia this commercial product became the first product in Russia which uses GPGPU technologies for performance enhancement. Leader of a team of developers who created the only industrial system in Russia for videoclassification of vehicles.
LinkedIn

Arlazarov Vladimir L.
Chief Science Officer

Ph.D., Professor, member-correspondent of the Russian Academy of Sciences. Research interests: system programming, game theory and artificial intelligence. V. L. Arlazarov is the author of classic method, which became known as “Method of Four Russians” (co-authors – E. A. Dinic, M. A. Kronrod and I.A. Faradzev). More detailed information can be found in Wikipedia.

Our key customers:

Tinkoff Bank uses mobile and server solutions for the documents recognition developed by Smart Engines.

Smart Engines completed a project to develop an automatic classification system for documents for PJSC “B&N BANK”.

Our customer is Jumio, one of world-leading developers of identification tools for payment service providers.

Smart Engines solutions for recognition of credit cards and MRZ elements of various documents are used by Facebook.

Smart Engines Ltd. supplies mobile OCR technology to S&T System Integration and Technology Distribution AG (S&T Romania).