Date: 4 months ago   Category: Hi-Tech

Release of the Tesseract 3.0 OCR system


The release of system of optical recognition of the text of Tesseract 4.0 supporting recognitions of documents in more than 100 languages including Russian and Ukrainian is published. The result can remain as clear text, and in HTML formats, PDF and TSV. Initially the system has been created in 1985-1995 in laboratory of the Hewlett Packard company, in 2005 the code has been open under the license Apache and further developed with the assistance of employees of the Google company. Source texts proyektarasprostranyatsya under the licenses Apache 2.0.

Main improvements in Tesseract 4.0: The new cursor of recognition which is based on use of system of machine learning on the basis of the recurrent neural network LSTM optimized for recognition entirely of lines and allowing to achieve essential increase in accuracy. The old cursor recognizing the text at the level of separate symbols continues to be delivered as an option. The alternative cursor of Cube is removed from code base; The utilities accompanying a new OCR cursor for a training of model of machine learning are added. The ready trained models are published for 123 languages; For a new cursor of recognition modules for optimization of productivity by means of OpenMP and SIMD instructions of AVX2, AVX or SSE4.1 are prepared; Transition to numbering of versions according to the Semantic Versionirovaniye specification meaning use of a notation of X.Y.Z Is carried out (instead of earlier used W.X.Y.Z) in which X changes at modification, breaking the return compatibility, Y changes at expansion of functionality and Z at correction of mistakes. The assembly system is updated. Assembly requires the compiler with support of C 11 now. Unit-tests are added. The assembly mode without old cursor of recognition is added. The reogranization of a tree of source texts is executed. The outdated code is purged;

Is improved processing of multipage TIFF images;

Is improved export to PDF. On materials: www.opennet.ru

URL:




Today

just now

Six matches of the regular championship were held in the National Basketball Association. Philadelphia 76 - Miami Heat - 106:102 (29:21, 28:29, 16:25, 33:27) MX: Wade - 19, Veyters - 18 of FS: H...

just now

Game Informer is not tired to tell about The Outer Worlds. Journalists devoted today's material to partners of whom they were told by screenwriters of Obsidian Entertainment. Though is also engaged...

just now

Unknown attacked and cruelly beat the known dogkhanter of Alexey Svyatogor near the Kiev Lukyanovskaya metro station. Is known that an incident occurred in the evening on Thursday, February 21, on t...

just now

The national bank Privatbank reported to clients that it for convenience of clients increased network of POS terminals for reception of non-cash and contactless payment of goods and services. It is sa...

just now

The Ukrainian singer acted in a revealing dress in Nalchik where the great number of Muslims lives. the Ukrainian singer of Ania Lorac Living in Russia ran into criticism from local Muslims. The sin...

just now

Apparently, the Kia company prepared for the Geneva motor show two very intriguing premieres at once. We already wrote about the first: it is the Xceed model, the pseudo-crossover on the basis of the...

just now

SpaceX carried out the second start of Falcon 9 this year, and it was successful again. Falcon 9 took the Indonesian communication satellite, the test apparatus of the Air Force and also the Israeli l...

just now

BenQ released the PD3220U monitor with a diagonal of 31.5 inches with a resolution of 3840? 2160. For it the 100 percent coverage of the sRGB color space and 95 percent coverage of space of DCI-P3 is...