man ocrodjvu (1): OCR for DjVu files

SYNOPSIS

ocrodjvu {-o | --save-bundled} output-djvu-file [option...] djvu-file
ocrodjvu {-i | --save-indirect} index-djvu-file [option...] djvu-file
ocrodjvu --save-script script-file [option...] djvu-file
ocrodjvu --in-place [option...] djvu-file
ocrodjvu --dry-run [option...] djvu-file
ocrodjvu {--version | --help | -h | --list-engines | --list-languages}

DESCRIPTION

ocrodjvu is a wrapper for OCR systems that allows you to perform OCR on DjVu files.

The following OCR engines are supported:

: • m[blue]OCRopusm[][1] (internally, ocrodjvu calls ocroscript's recognize (or rec-tess) command, so that ultimately Tesseract acts as the OCR backend);

: • m[blue]Cuneiform for Linuxm[][2].

: • m[blue]Ocradm[][3].

: • m[blue]GOCRm[][4].

: • Stand-alone m[blue]Tesseractm[][5].

OPTIONS

OCR engine options

-e, --engine=engine-id

: Use this OCR engine.
The default is "tesseract". (The default was "ocropus" prior to ocrodjvu 0.8.)

--list-engines

: Print list of available OCR engines.

Options controlling output

-o, --save-bundled=output-djvu-file

: Save OCR results as a bundled multi-page document into output-djvu-file.

-i, --save-indirect=index-djvu-file

: Save OCR results as an indirect multi-page document. Use index-djvu-file as the index file name; put the component files into the same directory. The directory must exist and be writable.

--save-script=script-file

: Save a djvused script with OCR results into script-file.

--in-place

: Save OCR results in place.
(Use this option to retain compatibility with ocrodjvu < 0.2.)

--dry-run

: Don't change any files, throw OCR results away.

It is mandatory to use exactly one of the above options.

--ocr-only

: If OCR results are to be saved to a separate document (-o/--save-bundled or -i/--save-indirect), save only the pages selected for OCR.
The default is to save all pages, even when the -p/--pages option is in effect.

--clear-text

: Remove existing hidden text if present in the pages not selected for OCR.
(Use this option to retain compatibility with ocrodjvu < 0.2.)

--save-raw-ocr=output-directory

: Save raw OCR results (typically in the hOCR format) into output-directory. The directory must exist and be writable.

--raw-ocr-filename-template=template

Specifies the file naming scheme for raw OCR results.

The template language uses the m[blue]Python string formatting syntaxm[][6]. The following fields are available:

page, page+N, page-N

: page number, optionally shifted by a number N

: page identifier

id-ext

: page identifier without file extension

The default template is "{id-ext}".

Text segmentation options

-t lines, --details lines

: Record location of every line. Don't record locations of particular words or characters.
This is the default for OCRopus 0.2. The option is ineffective with stand-alone Tesseract 2.0.

-t words, --details=words

Record location of every line and every word. Don't record locations of particular characters.

This is the default for most OCR engines.

This option is ineffective with OCRopus 0.2 and stand-alone Tesseract 2.0.

-t chars, --details=chars

: Record location of every line, every word and every character.
This option is ineffective with OCRopus 0.2 and stand-alone Tesseract 2.0.

--word-segmentation=simple

: Consider each non-empty sequence of non-whitespace characters a single word.
This is the default, despite being linguistically incorrect.

--word-segmentation=uax29

: Use the m[blue]Unicode Text Segmentationm[][7] algorithm to break lines into words.
This option breaks assumptions of some DjVu tools that words are separated by spaces, and therefore it is not recommended.

Other options

-l, --language=language-id

Set recognition language. language-id is typically an ISO 639-2/T three-letter code.

Tesseract ≥ 3.02 allows specifying multiple languages separated by "+" characters.

For OCRopus, the default is "eng" (English), unless the tesslanguage environment variable is set. For other OCR engines, the default is always "eng".

--list-languages

: Print list of available languages for the currently selected OCR engine.

--render=mask

: Render only masks of page images.
This is the default.

--render=foreground

: Render only foreground layers of page images.

--render=all

: Render all layers of page images.
This option is necessary to OCR DjVu files with invalid foreground/background separation.

-p, --pages=page-range

: Specifies pages to process. page-range is a comma-separated list of sub-ranges. Each sub-range is either a single page (e.g. 17) or a contiguous range of pages (e.g. 37-42). Pages are numbered from 1.
The default is to process all pages.

-j, --jobs=n

: Start up to n OCR processes.

--version

: Output version information and exit.

-h, --help

: Display help and exit.

Advanced options

-D, --debug

: To ease debugging, don't delete intermediate files.

-X key=value

: This option allows controlling some details of how ocrodjvu operates.

--on-error=abort

: Stop program execution when an exceptional situation (e.g., malformed output from the OCR engine, internal ocrodjvu error, etc.) occurs.
This is the default.

--on-error=resume

: Attempt to recover from exceptional situations.
This option is strongly discouraged.

--html5

: Use a m[blue]HTML5 parserm[][8], which is more robust but slower than the default parser.

EXIT STATUS

One of the following exit values can be returned by ocrodjvu:

: The program finished successfully.

: A fatal error occurred.

: The program recovered from an error (--on-error=resume).

ENVIRONMENT

The following environment variables affects ocrodjvu:

tesslanguage

: Recognition language for Tesseract.
(Use this variable is deprecated in favor of the --language option.)

TMPDIR

: ocrodjvu makes heavy use of temporary files. It will store them in a directory specified by this variable. The default is /tmp.

BUGS

Known bugs

Tesseract 3.00 is affected by a bug [9] making it produce invalid hOCR output in certain circumstances. ocrodjvu does not try recover form this fault (which couldn't be done reliably anyway) unless you pass the -X fix-html=1 option.

When using Tesseract ≥ 3.00, extracting bounding boxes of particular characters (which happens when either --details=chars or --word-segmentation=uax29) is inefficient. This is due to limitations of the Tesseract command-line interface.