Skip to content

A basic tool that extracts the structure from the PDF files of scientific articles.

License

Notifications You must be signed in to change notification settings

data-house/pdfact

 
 

Repository files navigation

PdfAct

PdfAct is a tool for selectively extracting text blocks from PDF files, optionally together with layout information like positional information, font information, or color information. For example, PdfAct allows you to extract only the important pieces of text from a PDF file (e.g., the title, the section headings and the body text paragraphs) and to choose the granularity in which the text should be extracted (e.g., characters, words, text lines or paragraphs). Further features are: (1) the detection of correct paragraph boundaries (which is important because paragraphs could be interrupted by for example, column- or page breaks, or figures, or tables, etc.) (2) the detection of the natural reading order of text blocks (which is of particular importance in case of multi-column layouts); (3) the translation of ligatures and characters with diacritical marks; (3) the detection of the semantic roles played by the text blocks in a PDF file; and (4) merging hyphenated words.

Disclaimer: Extracting text from PDF files is a suprisingly difficult task, read our publication "A Benchmark and Evaluation for Text Extraction from PDF" to learn why. Further, PdfAct is still in an early alpha state and is still under development. So please excuse bugs and use PdfAct on your own risk :-)

1. Checkout

Checkout the repository

git clone https://github.com/ad-freiburg/pdfact.git
cd pdfact

2. Run PdfAct

You have several options to run PdfAct.

2.1 Without Docker

Build the project with

mvn install

This will produce the two executables ./bin/pdfact.jar and ./bin/pdfact. Both files are logically equivalent, as they accept the exact same command line options, will run the exact same code in the background and will also produce the exact same extraction output for the same PDF file. Both files simply provide two different options to run PdfAct from the command line:

Option 1:

java -jar ./bin/pdfact.jar [--help] [<options>] <pdf-file> [<output-file>]

Option 2:

./bin/pdfact [--help] [<options>] <pdf-file> [<output-file>]

Note that the syntax of both commands is principally the same and that they differ only in their prefixes ("java -jar ./bin/pdfact.jar" and "./bin/pdfact"). Obviously, Option 2 is much shorter and thus a bit more convenient to write.

Example commands

In the following, you will find some example commands that will give you an intuition on how to use PdfAct in general. Note that, for the sake of convenience, the commands are written in the shorter notation of Option 2. If you prefer to use the notation of Option 1 instead, just replace the "./pdfact" prefixes in the commands below by "java -jar ./pdfact.jar".

(1) Extract all text blocks from a PDF file as plain text and print them to the console.

./pdfact foo.pdf

This extracts all text blocks from the PDF file foo.pdf as plain text and prints them to the console (each text block separated by a blank line).

(2) Extract all text blocks from a PDF file as plain text and write them to a file.

./pdfact foo.pdf output.txt

Instead of printing the extraction output to the console, you can also write it to a file by specifying a file path as the second argument. In the command above, the extraction output is written to output.txt.

(3) Create a visualization of the extracted text blocks.

./pdfact foo.pdf --visualize visualization.pdf

This creates the PDF file /home/user/visualization.pdf, which is in fact a copy foo.pdf with each extracted text block surrounded by its bounding box and each extracted paragraph labelled with its semantic role.

(4) Extract only selected text blocks, together with layout information.

./pdfact foo.pdf --format json --roles heading,body

This extracts the text blocks with the semantic roles "heading" and "body", together with their layout information (the positions in the PDF, the font information and the color information). The output is encoded in JSON format and provides the layout information in form of key-value pairs (TODO: Explain the exact format in detail). Instead of json you can also choose the format xml which will encode the exact same layout information in XML format (TODO: Explain this format as well).

(5) Print usage info.

./pdfact --help

The output and the behavior of PdfAct is highly customizable. To get a complete usage info with an overview of all available options, type the command above.

2.2 With Docker

If you want to use Docker instead, you first have to build the Docker image:

docker build -t pdfact .

The basic syntax of the Docker command to run PdfAct is as follows:

docker run --rm pdfact [--help] [<options>] <pdf-file> [<output-file>]

Example commands

In the following, we give the Docker commands equivalent to the example commands in Section 2.1: (1) Extract all text blocks from a PDF file as plain text and print them to the console.

docker run --rm -v /home/user/foo.pdf:/input.pdf pdfact input.pdf

(2) Write the extraction output to a file.

docker run --rm -v /home/user/foo.pdf:/input.pdf -v /home/user/output.txt:/output.txt pdfact input.pdf output.txt

NOTE: /home/user/output.txt must exist on the host before running the command, because otherwise Docker creates and mount the empty directory /home/user/output.txt.

(3) Create a visualization of the extracted text blocks.

docker run --rm -v /home/user/foo.pdf:/input.pdf -v /home/user/visualization.pdf:/visualization.pdf pdfact input.pdf --visualize visualization.pdf

NOTE: Similar to the note above, /home/user/visualization.pdf must exist on the host before running the command, because otherwise Docker creates and mount the empty directory /home/user/visualization.pdf.

(4) Extract selected text blocks together with layout information.

docker run --rm -v /home/user/foo.pdf:/input.pdf pdfact input.pdf --format json --roles heading,body

(5) Print usage info.

docker run --rm pdfact --help

About

A basic tool that extracts the structure from the PDF files of scientific articles.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Languages

  • Java 99.7%
  • Other 0.3%