formhtr 0.3.0

Handwritten text recognition in form documents.

formHTR

Handwritten text recognition in form documents.

Installation

pip

pip install formhtr

The tool also requires the zbar shared library installed (used by pyzbar). For PDF-related tooling, qpdf is also required.

System dependencies:

• macOS (Homebrew): brew install zbar qpdf
• Debian/Ubuntu: sudo apt-get install libzbar0 qpdf
• Fedora: sudo dnf install zbar qpdf

You can verify runtime requirements with:

formhtr doctor

Usage

Run formhtr --help for full CLI help.

Quickstart

# 1) Verify system dependencies
formhtr doctor

# 2) Create ROI config for a template
formhtr select-rois --pdf-file template.pdf --output-file config.json

# 3) Optionally annotate ROI types and variable names
formhtr annotate-rois --pdf-file template.pdf --config-file config.json --output-file config_annotated.json

# 4) Process a scanned logsheet into XLSX
formhtr process-logsheet \
  --pdf-logsheet scan.pdf \
  --pdf-template template.pdf \
  --config-file config_annotated.json \
  --output-file output.xlsx \
  --google google_credentials.json \
  --amazon amazon_credentials.json \
  --azure azure_credentials.json

Select ROIs

Find and define locations of regions of interest (ROIs) in the given PDF.

Generally, it is possible to draw ROIs (rectangles) manually but also to detect them automatically. The coordinates of ROIs are stored in a JSON file.

The tool is supposed to be run from the command line, as the control commands are entered there.

Control commands

• Press q or Esc to exit editing and save the config file.
• Press r to remove the last rectangle.

Run formhtr select-rois -h for details.

Annotate ROIs

Specify the type of content for each rectangle.

The workflow is designed in a way that you can navigate over specified ROIs and assign them the expected type of their content. This is done by pressing appropriate control commands.

Control commands

• Press q or Esc to exit editing and save the config file.
• Press h to add "Handwritten" type to the current ROI.
• Press c to add "Checkbox" type to the current ROI.
• Press b to add "Barcode" type to the current ROI.
• Press r or d to delete the type from the current ROI.
• Press v to enter the variable name.
• Press an arrow to navigate through ROIs (only left and right for now).

Run formhtr annotate-rois -h for details.

Process logsheet

Extract values from specified ROIs.

This is the crucial step that applies various techniques to extract the information as precisely as possible. It can process one logsheet at a time, given the template and config files.

Run formhtr process-logsheet -h for details.

Credentials

The processing of logsheets is using external services requiring credentials to use them. Here we specify structure that is expected for credentials, always in JSON format.

Google

{
  "type": "service_account",
  "project_id": "theid",,
  "private_key_id": "thekey",
  "private_key": "-----BEGIN PRIVATE KEY-----anotherkey-----END PRIVATE KEY-----\n"
  "client_email": "emailaddress",
  "client_id": "id",
  "auth_uri": "https://accounts.google.com/o/oauth2/auth",
  "token_uri": "https://oauth2.googleapis.com/token",
  "auth_provider_x509_cert_url": "https://www.googleapis.com/oauth2/v1/certs",
  "client_x509_cert_url": "someurl",
  "universe_domain": "googleapis.com"
}

Amazon

{
    "ACCESS_KEY": "YOUR_KEY_ID_HERE",
    "SECRET_KEY": "YOUR_ACCESS_KEY_HERE",
    "REGION": "YOUR_REGION_NAME_HERE"
}

Microsoft

{
    "SUBSCRIPTION_KEY": "YOURKEYHERE",
    "ENDPOINT": "https://ENDPOINT"
}

Examples

From the repository root (after git clone), set a short path prefix:

DATA=tests/test-data

CLI

# Check qpdf / zbar (no test files required)
formhtr doctor

# Automatic corner alignment payload for front (and optional back) pages
formhtr automatic-align \
  --pdf-logsheet "$DATA/logsheet/logsheet_tara.pdf" \
  --pdf-template "$DATA/template/template_tara.pdf"

# Two-page scan with separate front/back templates (CTD fixtures)
formhtr automatic-align \
  --pdf-logsheet "$DATA/logsheet/logsheet_ctd.pdf" \
  --pdf-template "$DATA/template/template_ctd_front.pdf" \
  --backside-template "$DATA/template/template_ctd_back.pdf"

# Export ROI crops to XLSX without OCR (use --no-aligned if the scan must be aligned first)
formhtr export-logsheet \
  --pdf-logsheet "$DATA/logsheet/logsheet_tara.pdf" \
  --pdf-template "$DATA/template/template_tara.pdf" \
  --config-file "$DATA/config/config_tara.json" \
  --output-file /tmp/formhtr_export_tara.xlsx \
  --aligned

# Full OCR pipeline (requires at least one of --google / --amazon / --azure)
formhtr process-logsheet \
  --pdf-logsheet "$DATA/logsheet/logsheet_tara.pdf" \
  --pdf-template "$DATA/template/template_tara.pdf" \
  --config-file "$DATA/config/config_tara.json" \
  --output-file /tmp/formhtr_out_tara.xlsx \
  --google credentials/google_credentials.json

# Two-sided CTD example (paths mirror tests)
formhtr process-logsheet \
  --pdf-logsheet "$DATA/logsheet/logsheet_ctd.pdf" \
  --pdf-template "$DATA/template/template_ctd_front.pdf" \
  --config-file "$DATA/config/config_ctd_front.json" \
  --output-file /tmp/formhtr_out_ctd.xlsx \
  --backside \
  --backside-template "$DATA/template/template_ctd_back.pdf" \
  --backside-config "$DATA/config/config_ctd_back.json" \
  --google credentials/google_credentials.json

# Interactive: click corner correspondences, write aligned PDF
formhtr manual-align \
  --pdf-template "$DATA/template/template_tara.pdf" \
  --pdf-logsheet "$DATA/logsheet/logsheet_tara.pdf" \
  --output /tmp/formhtr_aligned_tara.pdf

# Interactive GUI: set ROI types and variable names
formhtr annotate-rois \
  --pdf-file "$DATA/template/template_tara.pdf" \
  --config-file "$DATA/config/config_tara.json" \
  --output-file /tmp/config_annotated.json

Python API

Same paths as above; run with PYTHONPATH=src from the repo root, or after pip install formhtr with paths adjusted to your checkout.

from pathlib import Path

DATA = Path("tests/test-data")

# formhtr.deps
from formhtr.deps import check_system_dependencies, ensure_system_dependencies

check_system_dependencies()
ensure_system_dependencies({"qpdf", "zbar"})

# formhtr.pdf_utils.get_pdf_dimensions
from formhtr.pdf_utils import get_pdf_dimensions

get_pdf_dimensions(pdf_file=str(DATA / "template/template_tara.pdf"))

# formhtr.auto_align
from formhtr.auto_align import build_alignment_payload, get_page_alignment_data

get_page_alignment_data(
    scanned_logsheet_pdf=str(DATA / "logsheet/logsheet_tara.pdf"),
    template_pdf=str(DATA / "template/template_tara.pdf"),
    page=0,
)
build_alignment_payload(
    scanned_logsheet_pdf=str(DATA / "logsheet/logsheet_ctd.pdf"),
    template_pdf=str(DATA / "template/template_ctd_front.pdf"),
    backside_template_pdf=str(DATA / "template/template_ctd_back.pdf"),
)

# formhtr.export_logsheet.export_logsheet_to_xlsx
from formhtr.export_logsheet import export_logsheet_to_xlsx

export_logsheet_to_xlsx(
    scanned_logsheet_pdf=str(DATA / "logsheet/logsheet_tara.pdf"),
    template_pdf=str(DATA / "template/template_tara.pdf"),
    config_json=str(DATA / "config/config_tara.json"),
    output_xlsx="/tmp/formhtr_export_tara.xlsx",
    already_aligned=True,
)

# formhtr.logsheet — OCR needs real credentials (see Credentials below)
from formhtr.logsheet import (
    ServiceCredentials,
    extract_logsheet,
    load_credentials,
    process_logsheet_to_xlsx,
)

creds = load_credentials(google_credentials_path="credentials/google_credentials.json")
extract_logsheet(
    scanned_logsheet_pdf=str(DATA / "logsheet/logsheet_tara.pdf"),
    template_pdf=str(DATA / "template/template_tara.pdf"),
    config_json=str(DATA / "config/config_tara.json"),
    credentials=creds,
    debug=False,
    front=True,
)
process_logsheet_to_xlsx(
    scanned_logsheet_pdf=str(DATA / "logsheet/logsheet_tara.pdf"),
    template_pdf=str(DATA / "template/template_tara.pdf"),
    config_json=str(DATA / "config/config_tara.json"),
    output_xlsx="/tmp/formhtr_pipeline_tara.xlsx",
    credentials=creds,
)

# formhtr.roi_tools (GUIs unless select_rois(..., headless=True))
from formhtr.roi_tools import annotate_rois, select_rois

select_rois(
    template_pdf=str(DATA / "template/template_tara.pdf"),
    output_config_json="/tmp/rois.json",
    headless=True,
    existing_config_json=str(DATA / "config/config_tara.json"),
)
annotate_rois(
    template_pdf=str(DATA / "template/template_tara.pdf"),
    config_json=str(DATA / "config/config_tara.json"),
    output_config_json="/tmp/rois_annotated.json",
)

# formhtr.manual_align (opens windows unless template_points/target_points are passed)
from formhtr.manual_align import align_page, manual_align_pdf

manual_align_pdf(
    template_pdf=str(DATA / "template/template_tara.pdf"),
    scanned_logsheet_pdf=str(DATA / "logsheet/logsheet_tara.pdf"),
    output_pdf="/tmp/formhtr_aligned_tara.pdf",
)
# align_page is used on in-memory images inside the library; typical use is via manual_align_pdf.

Documentation

API reference and installation notes are built with Sphinx under docs/source/. After you connect this repository to Read the Docs (import the GitHub repo; the root .readthedocs.yaml drives the build), each push to the default branch triggers a documentation build. Enable “Build pull requests” in the RTD project settings if you want preview builds for PRs.

Pull requests also run a GitHub Actions job (.github/workflows/docs.yml) that installs the package and runs sphinx-build, so broken docs fail CI before merge.

Build locally (install Poppler / poppler-utils alongside zbar and qpdf so pdf2image can run):

pip install ".[docs]"
sphinx-build -b html docs/source docs/_build/html
# equivalent: make -C docs html

Developer Documentation

Setup

Create your development environment using the provided script via conda to install all required dependencies.

conda (dev)

conda env create -f conda_env.yaml

Contributing

We appreciate contributions - feel free to open an issue on our repository, create your own fork, work on the problem and post a PR. Please add your contributions to the changelog and to adhere to the versioning. For more information see here.

Testing

All functionality is tested with the pytest framework.

The repository includes unit and mocked integration-style tests for the OCR pipeline, CLI dispatch/validation, and output generation helpers. Tests are executed automatically on pull requests and on pushes to main.

Run locally:

python -m pip install -r requirements.txt
python -m pytest -q

For local development from a git checkout, convenience wrapper scripts live in scripts/ and forward to the CLI entrypoints (for example python scripts/process_logsheet.py ...).

Optional live OCR test:

• Add at least one credential file in credentials/: google_credentials.json, amazon_credentials.json, or azure_credentials.json.
• The live test is skipped automatically when credentials are missing and in CI.

python -m pytest -q -m live_services