release of the optical character recognition system , supporting UTF-8 character recognition and texts in over 100 languages, including Russian, Kazakh, Belarusian, and Ukrainian. The output can be saved as plain text or in HTML (hOCR), ALTO (XML), PDF, and TSV formats. The system was originally created between 1985 and 1995 in a laboratory at Hewlett Packard, and in 2005, the code was released under the Apache license and further developed with the participation of Google employees. The source texts of the project under the Apache 2.0 license.
Tesseract includes a command-line utility and the libtesseract library for integrating text recognition features into other applications. Among the third-party supporting Tesseract notable ones include , and . Two recognition engines are offered: a classic one that recognizes text at the character pattern level, and a new one based on the application of a machine learning system using an LSTM recurrent neural network, optimized for recognizing entire lines and allowing for a significant increase in accuracy. Pre-trained models have been published for . To optimize performance, modules utilizing OpenMP and AVX2, AVX, or SSE4.1 instruction sets are recommended.
Key In Tesseract 4.1:
- The capability to output in XML format has been added (Analyzed Layout and Text Object). To use this format, the application should be run as "tessaract image_name output_directory alto";
- New modules for rendering, LSTMBox and WordStrBox, have been added to simplify engine training;
- Support for pseudographics in hOCR (HTML) output has been added;
- Alternative scripts for training the machine learning-based engine written in Python have been added;
- Optimizations using AVX, AVX2, and SSE instructions have been expanded;
- OpenMP support is disabled by default due to performance concerns;
- Support for white and blacklists has been added in the LSTM engine;
- Build scripts based on CMake have been improved.
Source: opennet.ru
