Hieroglyph

Hieroglyph is the system of artificial intelligence intended for recognition of any certifying documents used in the world.

Hieroglyph are the cornerstone own scientific developments of the company in the field of visual intelligence. Deep integer neural networks are applied to solving of tasks of detection, classification and recognition of a large number of document types in Hieroglyph.

The technology of computationally effective visual memory in Hieroglyph, in turn, is based on creation for images of the compact descriptions invariant to luminance and space errors (including projective) in the form of vectors of multidimensional metric space. It is especially important for high-quality receiving recognition during the work with a video flow in mobile applications at unknown lighting.

The technology of synthesis of selections implemented in Hieroglyph for training of neural networks allows to add recognition of new document types, having only 2 images of this document type. One — for training of a system, and the second — for check of result.

2019: Recognition of hieroglyphs of China, Japan and Korea

On February 27, 2019 the Smart Engines company provided a sensing technology of hieroglyphs. Its development became possible thanks to application of the deep rarefied discrete neural networks which allow to distinguish over 45,000 hieroglyphs on documents with a difficult background and under different conditions of lighting. From them 20,941 Chinese, 12,912 Japanese and 11,172 Korean hieroglyphs.

As reported in Smart Engines, the provided functionality is integrated into the system of artificial intelligence Hieroglyph which at the level of the platform supports recognition 800+ of types of certifying documents in more than 40 languages. These are ID cards, passports, driver's licenses, visas and other documents more than 170 countries of the world. Recognition of hieroglyphic writing allowed to expand potential user audience of the products Smart Engines more than on 1.5 billion people.

Recognition of hieroglyphs is a difficult scientific and technical task in which developers of artificial intelligence deal in about tens of thousands of characters, many of which slightly differ from each other. In development process specialists of Smart Engines could break an important technology barrier and pass from methods of the classical OCR expected character recognition on a white background on qualitative scans to the computer vision capable to distinguish hieroglyphs even in identity certificates in a video flow and in photos.

"The visible noises caused by presence on the image of a background in the document and the distortions caused by the different brightness of lighting and a perspective of shooting are added to the difficult device of hieroglyphs in our case. We managed to create special architecture of neural networks which provides high-speed performance and quality of recognition of hieroglyphs in documents at the level of characters of the latin alphabet and Cyrillics". Vladimir Arlazarov, CEO of Smart Engines PhD in Technological Sciences

According to the developer, the provided technology is held for use in processes of registration, identification and verification of users in mobile applications and Internet services of the banking sector, insurance, a fintekh, sheringovy economy, transport, a telecom, etc. According to information for March, 2019 to developers libraries of recognition of certifying documents of China, Hong Kong, Japan and Korea are available to iOS, to Android, Linux, Windows, macOS and Solaris. It is possible test recognition of the Chinese, Japanese and Korean hieroglyphs in a video flow and in photos having installed the free Smart IDReader demonstration application from Google play or App Store.

2018: Announcement of the Hieroglyph system

The Smart Engines company in March, 2018 provided the system of artificial intelligence Hieroglyph which allows to configure recognition of any certifying documents. Development of a technology platform of Hieroglyph was conducted during 2017, and, as of March 14, on it recognition of passports, ID cards and car driver licenses for more than 55 countries of the world is implemented. Directly at the level of the platform 38 languages are already supported, including practically all European, even such visually difficult as Armenian and Georgian.

According to the CEO of Smart Engines PhD in Technological Sciences of Vladimir Arlazarov, in terms of the world market of recognition countries of Western Europe and the USA are still of the greatest interest. Specialists of the company managed to localize a product for work with documents in the English, German, French, Spanish, Portuguese languages that allowed to expand potential geography of the solution to 1.5 billion users.

Recognition algorithms in Hieroglyph are optimized for architecture "Elbrus" SPARC ARM, x86 and are compatible with operating systems iOS Android Sailfish Mobile OS RUS, Linux (including Astra Linux and "Atliks") Windows, macOS and Solaris. The quality of recognition does not depend on a hardware-software environment, and high-speed performance is defined by available computing resources. Video, photos, scans of documents and their copies received using phones tablets, webcams can be processed and scanners.

Deep integer neuronets are applied to solving of tasks of detection, classification and recognition of a large number of document types in Hieroglyph. Scientists of the company managed to solve the known problem connected with big computing complexity of standard solutions in the field of training of machines. For optimization of high-speed performance of deep neural networks the separated architecture of convolution layers and 8-bit model of calculations was used. The majority of program functions of the platform support processing in 8-bit integer arithmetics that allows to calculate the reply of deep neural convolution networks in an eight-bit path.

Expects to receive the first significant results of sales of solutions based on Hieroglyph in international market of Smart Engines by the end of 2019.