Data extractor for PDF invoices - invoice2data

A command line tool and Python library to support your accounting process.

1. extracts text from PDF files using different techniques, like pdftotext, text, ocrmypdf, pdfminer, pdfplumber or OCR -- tesseract, or gvision (Google Cloud Vision).
2. searches for regex in the result using a YAML or JSON-based template system
3. saves results as CSV, JSON or XML or renames PDF files to match the content.

With the flexible template system you can:

• precisely match content PDF files
• plugins available to match line items and tables
• define static fields that are the same for every invoice
• define custom fields needed in your organisation or process
• have multiple regex per field (if layout or wording changes)
• define currency
• extract invoice-items using the lines-plugin developed by Holger Brunn

Go from PDF files to this:

{'date': (2014, 5, 7), 'invoice_number': '30064443', 'amount': 34.73, 'desc': 'Invoice 30064443 from QualityHosting', 'lines': [{'price': 42.0, 'desc': u'Small Business StandardExchange 2010\nGrundgeb\xfchr pro Einheit\nDienst: OUDJQ_office\n01.05.14-31.05.14\n', 'pos': u'7', 'qty': 1.0}]}
{'date': (2014, 6, 4), 'invoice_number': 'EUVINS1-OF5-DE-120725895', 'amount': 35.24, 'desc': 'Invoice EUVINS1-OF5-DE-120725895 from Amazon EU'}
{'date': (2014, 8, 3), 'invoice_number': '42183017', 'amount': 4.11, 'desc': 'Invoice 42183017 from Amazon Web Services'}
{'date': (2015, 1, 28), 'invoice_number': '12429647', 'amount': 101.0, 'desc': 'Invoice 12429647 from Envato'}

flowchart LR

    InvoiceFile[fa:fa-file-invoice Invoicefile\n\npdf\nimage\ntext] --> Input-module(Input Module\n\npdftotext\ntext\npdfminer\npdfplumber\ntesseract\ngvision)

    Input-module --> |Extracted Text| C{keyword\nmatching}

    Invoice-Templates[(fa:fa-file-lines Invoice Templates)] --> C{keyword\nmatching}

    C --> |Extracted Text + fa:fa-file-circle-check Template| E(Template Processing\n apply options from template\nremove accents, replaces etc...)

    E --> |Optimized String|Plugins&Parsers(Call plugins + parsers)

    subgraph Plugins&Parsers

      direction BT

        tables[fa:fa-table tables] ~~~ lines[fa:fa-grip-lines lines]

        lines ~~~ regex[fa:fa-code regex]

        regex ~~~ static[fa:fa-check static]

 

    end

    Plugins&Parsers --> |output| result[result\nfa:fa-file-csv,\njson,\nXML]

 

 click Invoice-Templates https://github.com/invoice-x/invoice2data/blob/master/TUTORIAL.md

 click result https://github.com/invoice-x/invoice2data#usage

 click Input-module https://github.com/invoice-x/invoice2data#installation-of-input-modules

 click E https://github.com/invoice-x/invoice2data/blob/master/TUTORIAL.md#options

 click tables https://github.com/invoice-x/invoice2data/blob/master/TUTORIAL.md#tables

 click lines https://github.com/invoice-x/invoice2data/blob/master/TUTORIAL.md#lines

 click regex https://github.com/invoice-x/invoice2data/blob/master/TUTORIAL.md#regex

 click static https://github.com/invoice-x/invoice2data/blob/master/TUTORIAL.md#parser-static

Loading

Installation

1. Install pdftotext

If possible get the latest xpdf/poppler-utils version. It's included with macOS Homebrew, Debian and Ubuntu. Without it, pdftotext won't parse tables in PDF correctly.

2. Install invoice2data using pip

   pip install invoice2data

Installation of input modules

An tesseract wrapper is included in auto language mode. It will test your input files against the languages installed on your system. To use it tesseract and imagemagick needs to be installed. tesseract supports multiple OCR engine modes. By default the available engine installed on the system will be used.

Languages: tesseract-ocr recognize more than 100 languages For Linux users, you can often find packages that provide language packs:

# Display a list of all Tesseract language packs
apt-cache search tesseract-ocr

# Debian/Ubuntu users
apt-get install tesseract-ocr-chi-sim  # Example: Install Chinese Simplified language pack

# Arch Linux users
pacman -S tesseract-data-eng tesseract-data-deu # Example: Install the English and German language packs

Usage

Basic usage. Process PDF files and write result to CSV.

• invoice2data invoice.pdf
• invoice2data invoice.txt
• invoice2data *.pdf

Choose any of the following input readers:

• pdftotext invoice2data --input-reader pdftotext invoice.pdf
• pdftotext invoice2data --input-reader text invoice.txt
• tesseract invoice2data --input-reader tesseract invoice.pdf
• pdfminer.six invoice2data --input-reader pdfminer invoice.pdf
• pdfplumber invoice2data --input-reader pdfplumber invoice.pdf
• ocrmypdf invoice2data --input-reader ocrmypdf invoice.pdf
• gvision invoice2data --input-reader gvision invoice.pdf (needs GOOGLE_APPLICATION_CREDENTIALS env var)

Choose any of the following output formats:

• csv invoice2data --output-format csv invoice.pdf
• json invoice2data --output-format json invoice.pdf
• xml invoice2data --output-format xml invoice.pdf

Save output file with custom name or a specific folder

invoice2data --output-format csv --output-name myinvoices/invoices.csv invoice.pdf

Note: You must specify the output-format in order to create output-name

Specify folder with yml templates. (e.g. your suppliers)

invoice2data --template-folder ACME-templates invoice.pdf

Only use your own templates and exclude built-ins

invoice2data --exclude-built-in-templates --template-folder ACME-templates invoice.pdf

Processes a folder of invoices and copies renamed invoices to new folder.

invoice2data --copy new_folder folder_with_invoices/*.pdf

Processes a single file and dumps whole file for debugging (useful when adding new templates in templates.py)

invoice2data --debug my_invoice.pdf

Recognize test invoices: invoice2data invoice2data/test/pdfs/* --debug

Use as Python Library

You can easily add invoice2data to your own Python scripts as library.

from invoice2data import extract_data
result = extract_data('path/to/my/file.pdf')

Using in-house templates

from invoice2data import extract_data
from invoice2data.extract.loader import read_templates

templates = read_templates('/path/to/your/templates/')
result = extract_data(filename, templates=templates)

Template system

See invoice2data/extract/templates for existing templates. Just extend the list to add your own. If deployed by a bigger organisation, there should be an interface to edit templates for new suppliers. 80-20 rule. For a short tutorial on how to add new templates, see TUTORIAL.md.

Templates are based on Yaml or JSON. They define one or more keywords to find the right template, one or more exclude_keywords to further narrow it down and regexp for fields to be extracted. They could also be a static value, like the full company name.

Template files are tried in alphabetical order.

We may extend them to feature options to be used during invoice processing.

Example:

issuer: Amazon Web Services, Inc.
keywords:
- Amazon Web Services
exclude_keywords:
- San Jose
fields:
  amount: TOTAL AMOUNT DUE ON.*\$(\d+\.\d+)
  amount_untaxed: TOTAL AMOUNT DUE ON.*\$(\d+\.\d+)
  date: Invoice Date:\s+([a-zA-Z]+ \d+ , \d+)
  invoice_number: Invoice Number:\s+(\d+)
  partner_name: (Amazon Web Services, Inc\.)
options:
  remove_whitespace: false
  currency: HKD
  date_formats:
    - '%d/%m/%Y'
lines:
    start: Detail
    end: \* May include estimated US sales tax
    first_line: ^    (?P<description>\w+.*)\$(?P<price_unit>\d+\.\d+)
    line: (.*)\$(\d+\.\d+)
    skip_line: Note
    last_line: VAT \*\*

The lines package has multiple settings:

• start > The pattern where the lines begin. This is typically the header row of the table. This row is not included in the line matching.
• end > The pattern denoting where the lines end. Typically some text at the very end or immediately below the table. Also not included in the line matching.
• first_line > Optional. This is the primary line item for each entry.
• line > If first_line is not provided, this will be used as the primary line pattern. If first_line is provided, this is the pattern for any sub-lines such as line item details.
• skip_line > Optional. If first_line is passed, this pattern indicates which sub-lines will be skipped and their data not recorded. This is useful if tables span multiple pages and you need to skip over page numbers or headers that appear mid-table.
• last_line > Optional. If first_line is passed, this pattern denotes the final line of the sub-lines and is included in the output data.

⚠️ Invoice2data uses a yaml templating system. The yaml templates are loaded with pyyaml which is a pure python implementation. (thus rather slow) As an alternative json templates can be used. Which are natively better supported by python.

The performance with yaml templates can be greatly increased 10x by using libyaml It can be installed on most distributions by: sudo apt-get libyaml-dev

Development

If you are interested in improving this project, have a look at our developer guide to get you started quickly.

Roadmap and open tasks

• integrate with online OCR?
• try to 'guess' parameters for new invoice formats.
• apply machine learning to guess new parameters / template creation
• Data cleanup per field
• advanced table parsing with pypdf_table_extraction

Maintainers

• Manuel Riel
• Alexis de Lattre
• bosd

Contributors

• Harshit Joshi: As Google Summer of Code student.
• Holger Brunn: Add support for parsing invoice items.

Used By

• Odoo, OCA module account_invoice_import_invoice2data

Related Projects

• OCR-Invoice (FOSS | C#)
• DeepLogic AI (Commercial | SaaS)
• Docparser (Commercial | Web Service)
• A-PDF (Commercial)
• PDFdeconstruct (Commercial)
• CVision (Commercial)