pdf-extract

A tool and library that can extract various areas of text from a PDF, especially a scholarly article PDF. It performs structural analysis to determine column bounds, headers, footers, sections, titles and so on. It can analyse and categorise sections into reference and non-reference sections and can split reference sections into individual references.

The latest version is 0.1.1. Earlier versions are far less reliable.

pdf-extract requires Ruby 1.9.1 or above.

Quick start

Install the latest version with:

$ gem install pdf-extract

Quick examples

Extract references from a PDF:

$ pdf-extract extract --references myfile.pdf

Extract references and a title from a PDF:

$ pdf-extract extract --references --titles myfile.pdf

Mark the locations of headers, footers and columns in a new PDF:

$ pdf-extract mark --columns --headers --footers myfile.pdf

Extract regions of text from a PDF, preserving line information (offsets from region origin):

$ pdf-extract extract --regions myfile.pdf

Extract regions of text from a PDF without line information (prettier and easier to read):

$ pdf-extract extract --regions --no-lines myfile.pdf

Resolve references to DOIs and output related metadata as BibTeX:

$ pdf-extract extract-bib --resolved_references myfile.pdf

Problems

pdf-extract mistakes normal text for references when attempting to extract references.

pdf-extract attempts to identify reference sections by comparing section features to an idealised model of a reference section. Sometimes this can go wrong. If pdf-extract is producing reference output that clearly includes something that is not a reference, try reducing the reference_flex slightly:

$ pdf-extract extract --references --set reference_flex:0.18 myfile.pdf

The default for reference_flex is 0.2. Make small decrements.

pdf-extract extracts no references.

As above, but try to increase the reference_flex a bit a time:

$ pdf-extract extract --references --set reference_flex:0.25 myfile.pdf

Keep trying with small increments to reference_flex. Note that a reference_flex of 1 means pdf-extract will identify all sections as reference sections.

pdf-extract is still producing weird output after fiddling with reference_flex.

Have a look at pdf-extract's settings:

$ pdf-extract settings

This command will produce a list of settings along with descriptions of what they affect. They can be set by passing a --set key:value argument to pdf-extract.