textract-hocr 0.1.3

Convert AWS Textract JSON output to hOCR format

textract-hocr

Convert AWS Textract JSON output to hOCR format for use with document processing tools.

Based on amazon-textract-hocr-output by AWS Samples.

Features

• ✅ Convert Textract JSON to hOCR HTML format
• ✅ hOCR 1.2 compliant* output
• ✅ Support for single and multi-page documents
• ✅ Basic Table extraction with full line/word structure
• ✅ Block grouping based on vertical/horizontal overlap (Limitation based on LTR and top-down reading order)
• ✅ Extract specific pages or page ranges from multi-page documents
• ✅ Automatic dimension detection from source images (PNG, JPEG, TIFF)
• ✅ Explicit dimension specification for PDFs (required)
• ✅ Force custom dimensions (override auto-detection)
• ✅ Fallback to Textract's default 1000x1000 dimensions
• ✅ Command-line interface and Python library
• ✅ Preserves text confidence scores and bounding boxes
• ✅ Configurable logging levels (info, warning, error)

• Note: hOCR spec is fairly loose in its requirements, and therefore there are a number of different interpretations and usages of hOCR type classes in OCR and PDF engines. This outputs similar elements to Tesseract, barring the usage of ocr_block instead of ocr_carea.

Installation

From PyPI (when published)

pip install textract-hocr

From source

git clone https://github.com/BlueBox-WorldWide/textract-hocr.git
cd textract-hocr
pip install -e .

Development installation

git clone https://github.com/BlueBox-WorldWide/textract-hocr.git
cd textract-hocr
pip install -e ".[dev]"

Usage

Command Line

Convert entire document:

textract-to-hocr input.json output.html

Convert with source image for automatic dimension detection:

textract-to-hocr input.json output.html --source image.png

Convert PDF with explicit dimensions (required for PDFs):

# For A4 at 300 DPI (8.27" x 11.69")
textract-to-hocr input.json output.html --width 2480 --height 3507

Convert specific page only:

textract-to-hocr input.json output.html --first-page 2 --last-page 2

Convert page range:

textract-to-hocr input.json output.html --first-page 2 --last-page 5

Convert from page 3 to end:

textract-to-hocr input.json output.html --first-page 3

Force specific dimensions (override auto-detection):

textract-to-hocr input.json output.html --width 2550 --height 3300

Control logging verbosity:

# Verbose output (info level)
textract-to-hocr input.json output.html --log-level info

# Default (warnings only)
textract-to-hocr input.json output.html --log-level warning

# Quiet (errors only)
textract-to-hocr input.json output.html --log-level error

Python Library

Convert entire document

from textract_hocr import textract_to_hocr
import json

# Load Textract JSON output
with open('textract_output.json', 'r') as f:
    textract_result = json.load(f)

# Convert to hOCR
hocr_html = textract_to_hocr(textract_result)

# Save to file
with open('output.html', 'w', encoding='utf-8') as f:
    f.write(hocr_html)

Convert with source image for automatic dimension detection

from textract_hocr import textract_to_hocr
import json

with open('textract_output.json', 'r') as f:
    textract_result = json.load(f)

# Provide source image path for auto-detection
hocr_html = textract_to_hocr(textract_result, source_file='scan.png')

with open('output.html', 'w', encoding='utf-8') as f:
    f.write(hocr_html)

Convert PDF with explicit dimensions (required)

from textract_hocr import textract_to_hocr
import json

with open('textract_output.json', 'r') as f:
    textract_result = json.load(f)

# For PDFs, you MUST provide explicit dimensions matching Textract's rasterization
# Example: A4 at 300 DPI (8.27" x 11.69")
hocr_html = textract_to_hocr(
    textract_result,
    dimensions={'width': 2480, 'height': 3507}
)

with open('output.html', 'w', encoding='utf-8') as f:
    f.write(hocr_html)

Convert specific page

from textract_hocr import textract_to_hocr
import json

with open('textract_output.json', 'r') as f:
    textract_result = json.load(f)

# Extract page 2 only (with explicit dimensions for PDF)
hocr_html = textract_to_hocr(
    textract_result, 
    first_page=2,
    last_page=2,
    dimensions={'width': 2480, 'height': 3507}  # Required for PDFs
)

with open('page2.html', 'w', encoding='utf-8') as f:
    f.write(hocr_html)

Convert page range

from textract_hocr import textract_to_hocr
import json

with open('textract_output.json', 'r') as f:
    textract_result = json.load(f)

# Extract pages 3-5 (with explicit dimensions for PDF)
hocr_html = textract_to_hocr(
    textract_result,
    first_page=3,
    last_page=5,
    dimensions={'width': 2550, 'height': 3300}  # Letter at 300 DPI
)

with open('pages_3_5.html', 'w', encoding='utf-8') as f:
    f.write(hocr_html)

Force custom dimensions

from textract_hocr import textract_to_hocr
import json

with open('textract_output.json', 'r') as f:
    textract_result = json.load(f)

# Override dimension detection
hocr_html = textract_to_hocr(
    textract_result,
    dimensions={'width': 2550, 'height': 3300}
)

with open('output.html', 'w', encoding='utf-8') as f:
    f.write(hocr_html)

Get document dimensions

from textract_hocr import get_document_dimensions

# From image (auto-detected)
dims = get_document_dimensions('image.png')
print(f"Width: {dims['width']}, Height: {dims['height']}")

# For PDFs, you MUST provide explicit dimensions
# This will raise ValueError:
# dims = get_document_dimensions('document.pdf')  # ERROR!

# Instead, provide dimensions explicitly:
dims = get_document_dimensions(
    'document.pdf',
    dimensions={'width': 2480, 'height': 3507}
)

# Or use dimensions parameter alone
dims = get_document_dimensions(dimensions={'width': 2550, 'height': 3300})

# Fallback to Textract defaults
dims = get_document_dimensions()  # Returns {'width': 1000, 'height': 1000}

What is hOCR?

hOCR is an open standard for representing OCR results in HTML format. It embeds text content along with layout information (bounding boxes, confidence scores, etc.) that can be used by document processing tools.

The hOCR format is widely supported by:

• Tesseract OCR
• OCRopus
• ABBYY FineReader
• Document analysis tools
• PDF overlay generators

Dimension Handling

The converter handles document dimensions in the following priority order:

1. Explicit dimensions (via dimensions parameter): Uses provided width/height
2. Image files (PNG, JPEG, TIFF, etc.): Auto-extracts actual pixel dimensions
3. PDF files: CANNOT auto-extract - you MUST provide explicit dimensions parameter
4. Fallback: Uses Textract's default 1000×1000 normalized dimensions

Why PDFs Require Explicit Dimensions

Textract rasterizes PDFs at a specific DPI (typically 200-300) before processing. The original PDF dimensions don't reliably indicate the resolution Textract used. Therefore, you must provide the dimensions matching Textract's rasterization:

• A4 at 300 DPI: {'width': 2480, 'height': 3507} (8.27" × 11.69")
• Letter at 300 DPI: {'width': 2550, 'height': 3300} (8.5" × 11")
• A4 at 200 DPI: {'width': 1654, 'height': 2339}
• Letter at 200 DPI: {'width': 1700, 'height': 2200}

Textract returns normalized coordinates (0-1 range). This tool converts them to pixel coordinates using the actual document dimensions for accuracy.

Output Format

The generated hOCR HTML includes:

• hOCR 1.2 compliant structure with proper DOCTYPE and metadata
• ocr_page divs with page dimensions
• ocr_block divs grouping lines with overlapping vertical positions
• ocr_table divs for tables with complete line and word structure
• ocr_line spans for text lines
• ocrx_word spans for individual words
• Bounding boxes in bbox left top right bottom format
• Confidence scores in x_wconf property
• Proper baseline information for line elements
• Content ordered by vertical position (top to bottom on page)

Block Grouping

Lines are grouped into ocr_block divs based on vertical overlap:

• Lines with overlapping Y-axis positions are grouped together
• Creates natural paragraph-like blocks without explicit paragraph detection
• Blocks use synthetic IDs (e.g., block_1_1, block_2_1)
• Each block's bounding box encompasses all contained lines

Table Support

Tables detected by Textract are converted to float div elements with ocr_table class:

• ocr_table rendered as <div> float elements (no HTML table structure)
• Each cell's content rendered as ocr_line spans containing ocrx_word spans
• Cell content in reading order (row by row, left to right)
• Bounding box and confidence score for the table region

Requirements

• Python 3.8+
• yattag >= 1.14.0
• Pillow >= 9.0.0 (for image dimension extraction)

License

MIT License - see LICENSE file for details.

Based on amazon-textract-hocr-output by AWS Samples.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add some amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Related Projects

• aws-samples/amazon-textract-hocr-output - Original implementation
• AWS Textract - AWS OCR service
• hOCR 1.2 Spec - hOCR 1.2 spec documentation
• Tesseract OCR - Popular open-source OCR engine with hOCR support

Support

If you encounter any issues or have questions:

1. Check existing GitHub Issues
2. Create a new issue with:
   • Your Python version
   • The error message or unexpected behavior
   • Sample input (if possible)
   • Steps to reproduce

Changelog

0.1.3 (2026-01-08)

Bug Fixes:

• Fixed reading order for lines on the same visual line
• Lines are now correctly grouped by vertical overlap and sorted left-to-right within each group
• Prevents incorrect grouping of non-overlapping lines

0.1.2 (2026-01-05)

Improvements:

• Added paragraph grouping (ocr_par)
• Improved intersection calculations (limited to LTR languages)

0.1.1 (2026-01-04)

Breaking Changes:

• PDF dimension handling changed: PDFs now require explicit dimensions parameter. Auto-extraction from PDF files has been removed due to reliability issues with determining Textract's rasterization DPI.
• Attempting to process a PDF without providing dimensions will now raise a ValueError with clear instructions.

Improvements:

• Added comprehensive logging throughout the conversion process
• Better error messages with actionable guidance for PDF dimension requirements
• Improved documentation with detailed examples for PDF processing at different DPIs
• Clearer function docstrings with examples for both image and PDF workflows

Dependency Changes:

• Removed PyPDF2 dependency (no longer needed)

Migration Guide: If you were using PDFs with auto-detection:

# Old (v0.1.0) - no longer works
hocr = textract_to_hocr(data, source_file='document.pdf')

# New (v0.1.1) - provide explicit dimensions
hocr = textract_to_hocr(
    data,
    dimensions={'width': 2480, 'height': 3507}  # A4 at 300 DPI
)

0.1.0 (2026-01-04)

• Initial release
• Support for single and multi-page conversion
• Image dimension auto-detection (PNG, JPEG, TIFF)
• PDF dimension extraction (removed in 0.1.1)
• Command-line interface
• Python library API
• Textract default dimension fallback
• Block grouping based on vertical overlap