Smart Text Engine. Recognition of Text Fragments
The Smart Text Engine program is designed for recognition of unstructured text fragments on documents and images.
Text fragment – is an image area containing a part of a text string. Text fragments are always single-line, but it is not regulated where they begin and where they end. Splitting of the actually printed text string into fragments is arbitrary.
State registration
The Smart ID Engine program is registered in the State Register of Computer Programs. The certificate of state registration No 2024611596. [1]
The program is based on technologies for searching text fragments, character recognition, etc., developed by Smart Engines.
You can set a set of acceptable “languages” basing on the system configuration and get the recognition result at the level of each text fragment both in the “raw” format and filtered by the selected language.
For detailed description of the here.
The use of the latest GreenOCR® technologies and artificial intelligence developed by our company ensures high recognition confidence. In order to improve quality, we have created ultra-compact, low-bit artificial intelligence models and are continuously improving them, as well as the product itself.
Due to the modular structure of the program, the size of the library is not large. That makes the library suitable for integration with variety of mobile and server systems.
The results are presented in a structured form within the framework of the SDK Smart Text Engine interface. Export to JSON is possible.
Delivery Package
- API in C, С++, C#, Java, Python, PHP, Javascript/webAssembly;
- samples in C/C++/C#/Java/PHP/Python/Javascript;
- frameworks: React Native, Flutter.
Advantages of Smart Text Engine
Information Security
No risk of data leakage. All recognition operations are fully automatic and do not impy the human factor. Recognition is performed offline on mobile phones or in the customer's network without manual verification or any operator intervention.
No network connection is required to work with our product. All processing is done in the local RAM of the devices (100% on-premise).
Recognition Quality and Speed
High confidence of recognition of text fields.
A good result is possible even at a low image resolution (from 100 DPI) and in uneven lighting, in bright light and in the dark, at any shooting angle.
Integration Possibilities
A wide range of operating systems is supported: Aurora OS, Android, iOS, MS Windows, Linux, Elbrus OS, RED OS, Astra Linux, Atlix OS, Alt Linux Aurora OS, х86, x86/64, ARMv7-v8-v9 (AArch32 and AArch64), MIPS (MIPS32 and MIPS64), Elbrus.
No strict requirements for the input documents
High resistance to the quality of the original image.
Text recognition is performed on any document, regardless of its type.
Functionality of Smart Text Engine
- Optional pre-search and rectification of text in an image;
- Recognition of unstructured text fragments on documents and in arbitrary images;
- Optical text recognition of documents with any format;
- Inverted text recognition;
- Full-text recognition;
- Recognition of the text both in the image of the document as a whole and in the images of individual fragments of documents;
- Recognition of typescript text fields;
- Font-independent recognition of printed text in Russian and English;
- Obtaining the recognition result both in the “raw” format and filtered by the selected language at the level of each text fragment. For example, if the image is recognized with the alphabet of the symbols of the languages “Rus + eng+ numbers”, then the recognition result of each fragment can be obtained both “as is” and for each language: “rus”, “eng”, “digits”;
- Providing coordinates of text objects in the original image;
- Assessment of recognition confidence at the level of characters, words or strings;;
- Text recognition without using ROM - without saving temporary and intermediate data on ROM;
- Information recognition without accessing resources external to the device (application servers, online services, etc.);
- The use of 8-bit and 4-bit deep neural network architectures for recognition of text and graphic elements;
- Control of the used computing threads number.
An image is processed in three modes:
- line - the local zone mode. Assumes that there is only one text fragment in the image.
- page - the document page mode. Preliminary search and correction of the rectangle of the document page.
- standard - the standard mode. Assumes that the text can be located on the image in any way.
By default, the standard mode is enabled.
The certificate of state registration No 2024611596. Included in the Unified Register of Russian programs for electronic computers and databases/ (entry in the register No 926023 from 28.12.2024) https://reestr.digital.gov.ru/search/?q=2024611596. ↩︎