Tesseract (software)
| Company type | Free software |
|---|---|
| Industry | Optical character recognition software |
| Predecessor | Proprietary software by Hewlett-Packard |
| Founded | 1985 |
| Founder | Hewlett-Packard |
| Headquarters | United States, United Kingdom |
Area served | Worldwide |
| Products | Optical character recognition engine |
| Owner | Google (development support since 2006) |
Tesseract is an optical character recognition (OCR) engine compatible with various operating systems. It is distributed as free software under the terms of the Apache License. Initially created by Hewlett-Packard in the 1980s as proprietary technology, it was made open source in 2005, with development support commencing from Google in 2006.
By the year 2006, Tesseract had established itself as one of the most precise open-source OCR engines available. Ongoing enhancements and backing from Google have solidified its status as a prominent OCR solution within the industry.
History
The Tesseract OCR engine originated as proprietary software developed at Hewlett-Packard laboratories located in Bristol, England and Greeley, Colorado, United States between 1985 and 1994. Updates were released in 1996 for Windows platforms, and a partial transition from C to C++ was completed by 1998. Initially, the majority of the codebase was written in C before being fully rewritten in C++. After a decade of limited activity, Hewlett-Packard and the University of Nevada, Las Vegas (UNLV) made the engine available as open-source software in 2005. Development sponsorship by Google began in 2006.
Version 4 introduced an OCR engine based on LSTM networks and broadened language compatibility to include 116 languages and 37 writing systems. Since 2018, contributions have been made by the Mannheim University Library through initiatives funded by the German Research Foundation. After more than two years of development, Version 5 was released in 2021.
Features
In 1995, Tesseract ranked among the top three OCR engines for character recognition accuracy. It supports operating systems such as Linux, Windows, and Mac OS X. Early iterations (up to version 2) only processed TIFF images containing simple single-column text and lacked layout analysis capabilities, resulting in suboptimal performance on complex documents. Starting with version 3, Tesseract incorporated page layout analysis, hOCR output format support, and compatibility with multiple image formats through integration with the Leptonica library. The engine can distinguish between monospaced and proportional fonts.
Language support initially covered only English but expanded to six Western languages by version 2. Version 3 further extended support to include ideographic languages such as Chinese and Japanese, right-to-left scripts like Arabic and Hebrew, among many others. Version 3.04 added an additional 39 languages and scripts, bringing total language support to over 100. Users can train Tesseract to recognize additional languages beyond those pre-included.
The engine effectively processes right-to-left text and numerous Indic scripts. When paired with frontends like OCRopus, it serves well as a backend OCR engine for complex recognition tasks. To achieve optimal accuracy, input preprocessing is recommended: scaling text height to a minimum of 20 pixels, correcting any rotation or skew present, applying high-pass filters to reduce brightness inconsistencies, and eliminating dark borders that might cause errors during binarization.
User interfaces
Tesseract functions primarily through a command-line interface. Although it does not include a native graphical user interface (GUI), several third-party projects provide GUIs that improve accessibility. Prominent examples are OCRFeeder and gImageReader, an open-source GUI available across multiple platforms.
Reception
In July 2007, Anthony Kay writing for Linux Journal characterized Tesseract as "a quirky command-line tool that does an outstanding job," emphasizing its superior text recognition capabilities compared to other open-source OCR engines despite having a limited feature set. In November 2020, Brewster Kahle from the Internet Archive commended Tesseract’s progress, noting that its accuracy had reached parity with proprietary OCR products and could be further tailored for specialized uses due to its modernized architecture.