Post by Luc Vincent, Uber Tech
LeadWe wanted to let you all know that a few months ago
we quietly released - or actually
re-released - an Optical
Character Recognition (OCR) engine into open source. You might wonder why Google is interested
in OCR? In a nutshell, we are all about making information available to users, and when this
information is in a paper document, OCR is the process by which we can convert the pages of
this document into text that can then be used for indexing.
This
particular OCR engine, called
Tesseract, was in fact
not originally developed at Google! It was developed at Hewlett Packard Laboratories between
1985 and 1995. In 1995 it was one of the top 3 performers at the OCR accuracy contest
organized by
University of Nevada in Las Vegas.
However, shortly thereafter, HP decided to get out of the OCR business and Tesseract has been
collecting dust in an HP warehouse ever since. Fortunately some of our esteemed HP colleagues
realized a year or two ago that rather than sit on this engine, it would be better for the
world if they brought it back to life by open sourcing it, with the help of the
Information Science Research Institute at UNLV.
UNLV was happy to oblige, but they in turn asked for our help in fixing a few bugs that had
crept in since 1995 (ever heard of bit rot?)... We tracked down the most obvious ones and
decided a couple of months ago that
Tesseract OCR was stable
enough to be re-released as open source.
A few things to know about
Tesseract OCR: for
now it only supports the English language, and does not include a page layout analysis module
(yet), so it will perform poorly on multi-column material. It also doesn't do well on
grayscale and color documents, and it's not nearly as accurate as some of the best commercial
OCR packages out there. Yet, as far as we know, despite its shortcomings, Tesseract is far
more accurate than any other Open Source OCR package out there. If you know of one that is
more accurate, please do tell us!
We are grateful to all the people at
HP who made it possible to release Tesseract into open source, and especially John Burns, who
championed and babysat the project. We would also like to thank the original Tesseract
development team, a partial list of whom is
here.
Last but not least, many thanks to our friends at
UNLV's ISRI, including Tom Nartker, Kazem
Taghva, Julie Borsack and Steve Lumos, for all their help with this project.
By the way, we are also hiring top-notch OCR engineers! See
this job posting for
more information.
