Smart Document Engine (ранее Smart DocumentReader)

Smart Document Engine (formerly Smart DocumentReader) is a system for automatic classification, recognition and extraction of details from structured, semi-structured and unstructured documents: help in the form of 2-NDFL, balance sheet, payment order and others.

2024

Tablet compatibility Kvadra_T

Smart Engines confirmed the compatibility of its software products and domestic Kvadra T tablets ones following a series of tests. Earlier, the company Yadro announced its intention to occupy 10-15% of PCs Russian market tablet in the middle price segment. Compatibility has been confirmed with the Smart Document Engine. Smart Engines reported this on August 9, 2024. More. here

Integration with Aurora OS

Smart Document Engine and other systems developed by the Russian company Smart Engines have become available to owners of devices based on the Aurora OS right in the web browser. This was announced on April 12, 2024 by representatives of Smart Engines. Read more here.

Red OS Compatibility 8

Smart Engines has tested the compatibility of the latest version of its technologies with Red OS 8. The company announced this on March 28, 2024.

Previously, the company's software products - Smart ID Engine, Smart Code Engine, Smart Document Engine and Smart Tomo Engine - received the 2.3.0 update. Read more here.

Smart Code Engine 2.3.0: Double the speed of MRZ recognition

Smart Engines on April 5, 2024 introduced an updated version of 2.3.0 software products.

The Smart Code Engine recognizes MRZ twice as quickly - a machine-readable area in passports or other identity cards containing basic information about the owner of the document. Smart Code Engine 2.3.0 is trained to extract data from a machine-readable area even in the most difficult cases: for example, if the MRZ on a document is curved, that is, it looks not like a straight line, but like an arc.

The upgrade also received a product for instant payments Universal Pay, which is part of the Smart Code Engine. This payment tool allows customers to automatically run the desired payment script when the camera of a mobile device hovers over a barcode, bank card or phone number. In this version, it can recognize phone numbers and card numbers in the same frame at the same time. Other improvements include facial account recognition in payment details and REST API support. Read more here.

Creating AI capable of detecting fake powers of attorney

Smart Engines Scientists trained to AI find traces of manipulations carried out with powers of attorney, court orders and other official documents. The developer announced this on February 26, 2024. Now it has become as easy to check any document as a general civil passport. RUSSIAN FEDERATION According to experts, an AI solution patented algorithms with will help reduce the number of frauds Russia with fake documents and identify traces of organized crime groups.

𝓲

Smart Engines

As representatives of Smart Engines specified to TAdviser, AI is built into the Smart Document Engine system. In AI training, the Smart Engines team used a one-shot learning approach. This means that to add a new document template, artificial intelligence does not need to train on a hundred examples and show all kinds of fakes. Algorithms need only one image (standard) of the document, and then they themselves will find anomalies - if any - on other samples. Smart Engines researchers have already patented algorithms and mechanisms used in the software in the Russian Federation. In 2023 alone, the company received 12 patents for inventions in the field of document authentication.

The hardware and software complex, which includes Smart Engines software and a special multispectral scanner, solves two key tasks at the same time. The system instantly extracts data from the document and checks it for authenticity and validity in three ranges - optical, ultraviolet (UV), infrared (IR). The entire scanning and verification process takes up to 10 seconds.

In recent years, the production of fake documents has reached a new level: organized criminal groups make fakes of such quality that a person is not able to identify by eye. Due to multispectrality and the latest advances in artificial intelligence, it has become possible to prevent the widest range of fraudulent attacks. And the number of these attacks is growing all the time: in 2023, the number of criminal cases in which forged documents appear exceeded 1.5 thousand.

The built-in AI checks the originality and security of the document form (protective fluorescent fibers, fluorescent ink, special printing paper, etc.), detects mechanical interventions in the document itself (mechanical data corrections, re-labels, and reprints, etc.), thereby fulfilling the mandatory requirements of the regulators.

In addition to powers of attorney, the developed decision checks for authenticity and other documents of the state model of the size of the A4: court orders, acts of civil status, diplomas, TCP of the car, etc. AI-based systems will be able to use government agencies, banks, notaries, as well as HR services of any company.

The number of crimes using forged documents has been steadily growing in recent years, and this is not about passports or driver's licenses, but about government-issued A4 documents. Notarial power of attorney is one of the most often forged documents by cybercriminals. It is clear that there is a special service of the Federal Notary Chamber, with which you can find out whether there is a power of attorney or not. But you will not find information about the content of the power of attorney there. Our software allows you to solve this problem and establish whether any manipulations were made in the text of the document, - said the general director of Smart Engines, Doctor of Technical Sciences Vladimir Arlazarov.

2023

Availability on Abanking

The company Abanking entered into a partnership agreement with the developer of recognition systems. Smart Engines The goal of the partnership is to promote the development and promotion of technologies for and. bank industries fintech Smart Engines announced this on October 31, 2023.

Abanking customers using the platform in their IT ecosystem will have access to all the key solutions of the Smart Engines product line. Among them are the flagship product Smart ID Engine, Smart Code Engine and Smart Document Engine. Read more here.

Smart Document Engine 2.1.0

On July 20, 2023, AI-company Smart Engines presented an updated version of software products for recognizing bank cards, QR codes, Russian passports and other certification documents, as well as for scanning primary, accounting and business documents.

As reported, in Smart Document Engine 2.1.0, the speed of recognition of full-text documents increased by 5%. The process of scanning an individual's income statement (previously 2-NDFL) on processor an x86 architecture has become 1.4 times faster. More. here

Smart Document Engine 2.0

On May 23, 2023, Smart Engines announced the release of the next generation corporate document processing system. Smart Document Engine 2.0 recognizes and verifies the digital authenticity of documents used by private companies and government agencies. These are primary accounting, accounting and tax accounting documents, corporate and personnel documents, as well as standard questionnaires.

Smart Document Engine 2.0

According to the company, Smart Document Engine 2.0 supports the recognition of 60 types of documents used in the Russian Federation. In the updated version of the program, recognition of all fields, the delivery note TORG-12, is available. most of the fields of the USRN, consignment note, price agreement protocol and others. Now the program reads tables in the reconciliation report, KS-2 forms, MKh-1, MKh-3, OS4. The solution in the updated version recognizes electronic vehicle passport (ETS).

In general, Smart Document Engine 2.0 recognizes 78 different types of documents for a number of countries and states, including 60 documents for Russia, 13 documents for the United States, four for Armenia and one for Belarus. For all recognized documents in the updated version, the packaging of results in PDF/A format has been improved.

This is an important milestone for us, the updated system covers most of the needs for recognition of accounting and document management documents. Smart Document Engine 2.0 not only recognizes data, but also detects attempts at fraud with corporate documents. told Vladimir Arlazarov, CEO of Smart Engines Candidate of Technical Sciences

In Smart Document Engine 2.0, the detection rate of all lines on a full-text document has increased 2.5 times. The file recognition process began to be carried out 2 times faster. Mechanisms for filtering false recognition and post-processing of recognition results have been optimized. Artificial intelligence automatically corrects characters according to the language model.

Sometimes OCR it is quite difficult to determine whether the number 0 or the letter "O" is present in the document, and it may be wrong. But if the solution recognizes a field where only numbers should be, then thanks to postprocessing, it will automatically select the correct character. This feature appeared in Smart Document Engine 2.0. noted Vladimir Arlazarov

All Smart Document Engine technologies are created by Smart Engines scientists. This is a completely Russian software product, it is included in the register of Russian software and is presented on the marketplace of the Ministry of Digital Development "Russoft."

SDK Smart Document Engine 2.0 in 102 languages ​ ​ is available for integration through APIs into domestic operating systems, any server, mobile, desktop and web applications. It does not need to connect to third-party services, external resources and the Internet. All calculations are performed on the central processor of the device. Personal data remains under the control of the client.

Most system-forming credit institutions trust Smart Engines technologies - proprietary document recognition systems and QR codes for May 2023 are used by nine of the 13 largest banks.

2022

Next Generation Text Recognition Release

The Russian company Smart Engines announced on December 27, 2022, the release of the next generation text recognition system. She knows how to find a document in a photo or scan and recognize all text data in 102 languages. The solution is part of the Smart Document Engine product included in the register of Russian programs. The proprietary GreenOCR character recognition technology used in all company products allows you to achieve the highest accuracy even in low-quality photos. The software is designed to replace ABBYY products and solutions based on them offered in Russia, as well as other foreign software in the corporate and public sector. 

Full-text recognition is a key element of document entry in electronic document management, business process management, electronic archives, and RPA systems. The speed and accuracy of data extraction directly depends on the complexity and possible depth of automation of the document processing process. 

The developed system provides technological sovereignty, since the product for recognizing and processing images does not use code Open Source and foreign software components. smartphone It takes 3-4 seconds for the entire process from photo to text, which makes tablet the scanner unnecessary. In addition to recognition, the system automatically crops, smoothes folded documents and improves its image by turning it phone into an instant scanner. server In 32s solutions without nuclear HPC the use of GPUs, the speed of full-text recognition reaches 15 pages per second.

Earlier in such tasks, many companies relied on the OCR solutions of ABBYY, but at the beginning of 2022, after 30 years of work in the country, she unexpectedly left the Russian Federation, excluding a number of products from the register of Russian programs. This event has become a "black swan" for the domestic market, creating significant risks in the implementation of digital transformation projects. In December 2022, the company presented the last missing element and now in Russia there are all the technologies necessary for business to recognize documents, CEO of Smart Engines, Candidate of Technical Sciences Vladimir Arlazarov.

For developers and customers, text document recognition technology is available for embedding in server, mobile, desktop and web applications as part of the Smart Document Engine software product. The system functions without connecting to third-party services and external resources, does not require a GPU. Recognition does not require the presence of an Internet, all calculations are performed on the central processor of the device and do not require the use of video cards.

In addition to the usual languages ​ ​ based on Cyrillic and Latin alphabets, it recognizes, Arab,, Armenian,, Greek, and Georgian. Hebrew Chinese Korean Smart Japanese Document Engine supports,, and ALT Linux other Astra Linux Red OS families OS ,,,,,. Linux Windows macOS iOS Android OS Aurora

AI for mobile and streaming input of primary documents

On June 2, 2022, Smart Engines released a recognition system for primary accounting and financial documentation on mobile phones with quality that was previously only available using manual verification systems. The Smart Document Engine allows you to add automatic entry of complex structured documents into mobile applications by replacing a full-fledged input center. Product characteristics allow you to implement the concept of a mobile backofice, when employees scan and enter data of primary documents using a smartphone or tablet solving business problems in real time. 

Снимки экрана smartphone recognized multi-page invoice

According to the company, the updated version "out of the box" automatically classifies and recognizes invoices, TORG-12, PDA, consignment notes, certificates and invoices for payment. Digitizing documents with the Smart Document Engine enables you to enter information from documents and forms into an ERP system or any other accounting information system with the ability to verify completeness and cross-verify data within one set. Now, recognizing the primary document on a modern phone in a mobile application, depending on its type and complexity, takes 1-3 seconds per page. In server mode on a 32-core HPC without a GPU, the Smart Document Engine recognition speed for streaming scans in traditional input centers can be up to 600 pages per minute.

We develop recognition algorithms exclusively based on our own stack of AI technologies for training ultralight neural networks. By applying them, we were able to achieve that even a mobile phone is able to recognize in real time a stream of up to 30 pages per minute. Now employees can scan and extract data with a mobile phone not only in single input centers, where paper documents are centrally received for processing, but also directly when receiving documents from counterparties. told Vladimir Arlazarov, Candidate of Technical Sciences, CEO of Smart Engines

Smart Document Engine supports data recognition on scans and photos not only in mobile applications, but can also be used to replace server-side primary document recognition systems implemented on the basis of programs recalled from the Registry or developed by companies that have left Russia. In this case, developers can migrate from them to the Smart Document Engine using a convenient SDK, without changing the logic of the current input system. 

Smart Document Engine is included in the Register of Russian software and can run on operating systems of the Linux family (including domestic distributions), Windows, iOS, Android, Aurora and Elbrus OS. The system does not contain dynamically loaded components from other developers, and the system uses its proprietary GreenOCR printed text recognition technology to recognize characters. In the recognition process, the Smart Document Engine does not use verifiers from third-party collaboration services or crowdsourcing platforms to enter data. 

Delivery of the Smart Document Engine to integrate primary document recognition capabilities into customer infrastructure and applications includes a standalone SDK, API document recognition documentation, and integration examples for,, C,, C++,,, C# Java Python PHP Objective C and. Swift You can test primary document recognition by installing the free Smart Document Engine demo application available in or. App Store Google Play

Release of version 1.10.0 with improved document identification and recognition speed

Smart Engines, a representative of the Russian market for automatic document recognition systems, on March 15, 2022 announced the release of version 1.10.0 for the entire line of its products.

Key updates include:

• improved accuracy of recognition of text fields based on the Latin alphabet and Cyrillic alphabet;
• improved speed of identification and recognition of documents;
• Document types added.

The Smart Document Engine is based on its own OCR engine, which previously provided high accuracy in passport recognition, RUSSIAN FEDERATION which guarantees clarity when automatically recognizing business documents and forms.

Smart Engines recognition technology is the company's own development of Russian scientists and does not contain external components, and all the company's products are included in the Register of Russian Software, support domestic computing platforms and do not require GPU.

Smart Engines software solutions perform real-time document recognition directly on the input device (user or client loop, without transferring to data third-party services) and do not require a network connection, the presence of graphic processors or powerful computing resources on the customer's side, and the use of their own OCR algorithms AI and guarantee high accuracy and speed of data recognition.

Smart Engines solutions are available for integration into mobile, desktop and server applications, work autonomously, ensuring the security of personal and sensitive data processing and support, along with Windows, Linux, iOS, Android, all key Russian operating systems: Astra Linux, Alt Linux, RED OS, Elbrus OS, mobile OS Aurora, etc.

Read about the key changes in the Smart Code Engine and Smart ID Engine products here and here, respectively.

2021

Red OS Compatibility

The Russian developer RED SOFT and the research company Smart Engines have confirmed the correctness of the joint operation of the operating system RED OS and document recognition systems Smart ID Engine, Smart Code Engine, Smart Document Engine. This was announced on August 11, 2021 by the Red Soft company. Read more here.

Browser Recognition Technology View

On June 21, 2021, the company Smart Engines introduced industrial technologies for recognizing documents in, which browser do not imply the transfer of source, intermediate or reference data from a client device. This solution is suitable for personal devices, objects Internet of things () IoT and minimizes the risk of leaks images with passport data of customers via the Internet. The company's researchers solved the difficult scientific and technical task of developing algorithms AI a real-time mode for full recognition in a browser and offered an alternative to recognition services for users, developers and businesses.

Using Smart Engines software products, users can quickly extract data and fill out online forms, and images of their documents will not leave the perimeter of the browser installed on their device. The developers have received a tool that allows, without creating special applications, to implement document recognition on gadgets around a person, including smart IoT devices. For business, recognizing passports, other documents, bank cards and barcodes in the browser means developing remote customer service channels based on the principles of omnichannel without threats to privacy and security.

The browser is the most versatile interface between a person and a device connected to the Internet. The operation of the program in the browser is the ability to provide customers with uniform service standards, regardless of which device the user works with and which software environment is used on this device. Using Smart Engines technologies, the functionality of recognizing passport data, RUSSIAN FEDERATION driver's license, SNILS, bank card, 2-NDFL, accounting or QR code has become available in web applications for,,, mobile phones tablets laptops desktop computers, as well as,, TVs smartwatch devices in the system and any smart home other smart devices equipped with a camera and browser.

Reliable and fast operation of recognition algorithms in the browser is achieved through the use of proprietary GreenOCR technology, which is based on the results of scientific developments of Smart Engines researchers in the field of low-bit neural network architectures. The use of specialized computer vision algorithms and original integer 8- and 4-bit models of computations of neural network architectures for execution, as well as deep algorithmic and software optimization carried out by the company's engineers, made it possible to ensure high recognition speed in the browser.

To extract data, users can take photos or recognize a document in a video stream by calling the device's camera on a web page. Technically, for recognition using Smart Engines technologies, any web browser that supports WebAssembly technology and a camera with a resolution of at least 640x480 must be installed on the user's device. WebAssembly allows you to run program code directly on a web page and perform all calculations in a browser, while using the low-level optimization capabilities of the platform on which it is running.

{{quote "In 2015, we presented a solution for safe recognition of the passport of the Russian Federation in mobile applications in real time, which did not send images to services and worked on the user's phone or tablet. Now we are opening another chapter in document recognition on the Internet.

The AI algorithms we have developed allow us to safely recognize the Russian passport and other documents in web applications in real time. As with mobile applications, our products are completely autonomous, work directly in the user's browser and do not transfer images for processing to services based on machine learning and/or using manual input of verifiers. From a business point of view, data recognition in a browser is not only a matter of concern for the safety of client data, but also the ability to reduce the cost of developing cross-platform applications on the way to building a client service, which is based on omnichannel and future-proof approach, - comments on the CEO of Smart Engines Ph.D. Vladimir Arlazarov.}}

Our paradigm for working with personal and sensitive data on the Internet is designed not only to ensure safe interaction for consumers in already created web applications, but also to lay the foundation for the emergence of various digital service channels and cross-sales by connecting the world of IoT devices. Definitely, Smart Engines artificial intelligence technologies are fully ready for the challenges of the Internet of Things era, - comments Dmitry Nikolaev, technical director of Smart Engines, Ph.D.

Inclusion in the register of Russian software

Smart Engines software products for recognizing bank cards, barcodes and standard documents are included in the register of Russian software. The developer announced this on March 19, 2021.

Software Products Smart Code Engine and Smart Document Engine are included in the software class, which includes linguistic software and subroutine libraries (SDKs).

Smart Engines solutions are based on the company's researchers in the field of creating energy efficient architectures. neural networks Their use in the process machine learning algorithms and recognition made it possible to achieve high speed and quality of automatic extraction. data The Smart Document Engine and Smart Code Engine work autonomously and do not transfer images for processing to third-party services or third parties for manual input, which allows companies to ensure the security of personal processing and sensitive customer data in their applications and systems.

Smart Document Engine and Smart Code Engine tools provide multi-platform and allow developers to embed recognition of documents, bank cards and barcodes in programs written for: operating systems,, iOS Android Sailfish Mobile,,,,, MOS "Aurora" Linux OS, Windows,, macOS Elbrus RED OS Astra Linux Atlix OS, OS, etc. Alt Linux The hardware architectures Elbrus,,, and x86 are supported. SPARC MIPS ARM

The inclusion of Smart Document Engine and Smart Code Engine in the register of domestic software confirms their compliance with the established rules and requirements of Russian legislation.

2020

Smart Document Engine: Automatic extraction of data from standard documents, strict reporting forms

On November 18, 2020, Smart Engines introduced the next generation of passport recognition systems, other identity cards, bank cards, barcodes and documents with authentication and biometric verification capabilities. The company has become a single supplier of technologies for data extraction, authentication of documents with verification of "vitality" (document liveness detection) and signs of compromise (computational document forensics), face matching (face matching) for user verification. All products of this line: Smart ID Engine, Smart Code Engine and Smart Document Engine are developed in accordance with the principles of responsible AI and are designed to protect users and businesses from fraudulent actions with documents. Read more here.

According to the company, Smart Document Engine solves the problems of automatic extraction of data from standard forms of documents, forms of strict reporting, primary accounting, financial, tax, legal, notarial and other documents used in document management, various tests and questionnaires, on scans and photographs. The system allows you to automatically process single and multi-page documents with a fixed position of details, documents with a floating arrangement of blocks and details, unstructured text documents and blocks, tables, inscriptions or even individual lines and labels.

The software product allows not only to quickly recognize data from questionnaires, forms and documents, but also to check them for compliance with formalities. The Smart Document Engine can check if there is a signature, seal or logo, whether they are the correct color, whether they are in the right place in the document, and check that the inscriptions to be handwritten are indeed handwritten. In addition, during processing, it is possible to check the integrity and invariability of the form, document or part of it. Due to the use of second generation GreenOCR technology, the processing time of 1 page of a document A4 on AMD Ryzen 7 3700X is about 2 seconds.

In the boxed delivery version, the Smart Document Engine supports the recognition of help by 2-NDFL form, balance sheet form (OKUD 0710001), financial results report (OKUD 0710002), TIN certificate, and payment order (OKUD 0401060).

Implementation in Basic Documents

On March 3, 2020 Smart Engines , he announced that he, Financial Technology Center "Basis" (CFT "Basis") representing comprehensive digital solutions in the field, mortgage crediting had launched a service for automating the entry of data certificates in the form of 2-NDFL. The solution significantly reduces the risk of erroneous data entry and speeds up the processing of client documents by 2.5 times.

The Smart DocumentReader technology developed by Smart Engines is responsible for the automatic part of the service, and the process of verification and typing of documents is based on the own solution of the Basis CFT. Read more here.

2019: Smart DocumentReader View

On May 22, 2019, Smart Engines introduced the Smart DocumentReader system, which can recognize complex documents with tables in photos and scans, even on mobile devices, without overheating them. This technology is implemented on the basis of the Hieroglyph AI platform developed by Smart Engines specialists. The first document available for recognition in Smart DocumentReader was the 2-NDFL form help.

Now banks fintech can offer customers a different user experience when entering data from 2-NDFL mobile applications Internet in and services. To do this, just take a picture of the document or select/upload an existing photo. From the point of view of the software architecture, this functionality complements the capabilities of the company's product Smart IDReader in terms of recognizing passports RUSSIAN FEDERATION and other certification documents in the robotizations credit pipeline in financial institutions. Information from the 2-NDFL is used to evaluate borrowers when issuing mortgages and other financial products, as well as in the services for issuing a set of documents for obtaining tax deductions.

Smart DocumentReader allows you to configure data recognition on any complex structured documents. Their difference from the "identical to the light" documents is the lack of regulations that determine not only the composition of the details, but also their exact location on the form of the document. In general, these can be single-page and multi-page documents, including those with a table part up to A4 inclusive. The most common such documents are: Invoice, Invoice, Certificate, TTN, TORG12, PDA, Articles of Association, Contract, Invoice, Questionnaire, Applications and others.

Smart DocumentReader allows you to extract 2-NDFL from several tens to hundreds of attributes on the help, including all the data of the table part, even when it is placed on 2 pages. The algorithms used for computationally efficient visual memory allow you to correct projective distortions and achieve high recognition quality even in photos taken by users in various lighting. If the library is embedded in a mobile application, you can recognize documents in real time by performing all calculations autonomously on a mobile device without transferring data for processing to external services.

The entire algorithmic base of Smart DocumentReader, from image preprocessing methods to optical character recognition (OCR), is Smart Engines' own development. To solve the problems of detection, classification and recognition of documents, ultralight deep integer neural networks are used . To optimize the performance of neural networks at the level of the HIEROGLYPH platform, integer arithmetic is used. Calculating the response of deep neural convolutional networks in an 8-bit path and implemented software and hardware optimizations avoid overheating when recognizing 2-NDFL even on mid-range mobile phones.

Vladimir Arlazarov commented on the release of the decision:

[[:Шаблон:Quote 'author '= Vladimir Arlazarov, CEO of Smart Engines Ph.D.']]

Smart DocumentReader is a multi-platform solution and is a developer tool that allows you to integrate complex document recognition algorithms into mobile, server and desktop applications. The technology supports the hardware platforms Elbrus, Komdiv, SPARC, MIPS, ARM, x86 and is compatible with the operating systems Sailfish Mobile OS RUS (Aurora), iOS, Android, Elbrus, Linux, Windows, macOS, Solaris.

As of May 2019, the 2-NDFL help recognition functionality based on Smart DocumentReader is available for testing only to customers of the company using Smart IDReader credential recognition technology or Smart CardReader bank card recognition technology.