[MS] Add OCR layer service for embedded images and PDF scans (#1541)
* Add OCR test data and implement tests for various document formats - Created HTML file with multiple images for testing OCR extraction. - Added several PDF files with different layouts and image placements to validate OCR functionality. - Introduced PPTX files with complex layouts and images at various positions for comprehensive testing. - Included XLSX files with multiple images and complex layouts to ensure accurate OCR extraction. - Implemented a new test suite in `test_ocr.py` to validate OCR functionality across all document types, ensuring context preservation and accuracy. * Enhance OCR functionality and validation in document converters - Refactor image extraction and processing in PDF, PPTX, and XLSX converters for improved readability and consistency. - Implement detailed validation for OCR text positioning relative to surrounding text in test cases. - Introduce comprehensive tests for expected OCR results across various document types, ensuring no base64 images are present. - Improve error handling and logging for better debugging during OCR extraction. * Add support for scanned PDFs with full-page OCR fallback and implement tests * Bump version to 0.1.6b1 in __about__.py * Refactor OCR services to support LLM Vision, update README and tests accordingly * Add OCR-enabled converters and ensure consistent OCR format across document types * Refactor converters to improve import organization and enhance OCR functionality across DOCX, PDF, PPTX, and XLSX converters * Refactor exception imports for consistency across converters and tests * Fix OCR tests to match MockOCRService output and fix cross-platform file URI handling * Bump version to 0.1.6b1 in __about__.py * Skip DOCX/XLSX/PPTX OCR tests when optional dependencies are missing * Add comprehensive OCR test suite for various document formats - Introduced multiple test documents for PDF, DOCX, XLSX, and PPTX formats, covering scenarios with images at the start, middle, and end. - Implemented tests for complex layouts, multi-page documents, and documents with multiple images. - Created a new test script `test_ocr.py` to validate OCR functionality, ensuring context preservation and accurate text extraction. - Added expected OCR results for validation against ground truth. - Included tests for scanned documents to verify OCR fallback mechanisms. * Remove obsolete HTML test files and refactor test cases for file URIs and OCR format consistency - Deleted `html_image_start.html` and `html_multiple_images.html` as they are no longer needed. - Updated `test_file_uris` in `test_module_misc.py` to simplify assertions by removing unnecessary `url2pathname` usage. - Removed `test_ocr_format_consistency.py` as it is no longer relevant to the current testing framework. * Refactor OCR processing in PdfConverterWithOCR and enhance unit tests for multipage PDFs * Revert * Revert * Update REDMEs * Refactor import statements for consistency and improve formatting in converter and test files
This commit is contained in:
lesyk
@@ -9,7 +9,7 @@
|
||||
|
||||
> [!IMPORTANT]
|
||||
> Breaking changes between 0.0.1 to 0.1.0:
|
||||
> * Dependencies are now organized into optional feature-groups (further details below). Use `pip install 'markitdown[all]'` to have backward-compatible behavior.
|
||||
> * Dependencies are now organized into optional feature-groups (further details below). Use `pip install 'markitdown[all]'` to have backward-compatible behavior.
|
||||
> * convert\_stream() now requires a binary file-like object (e.g., a file opened in binary mode, or an io.BytesIO object). This is a breaking change from the previous version, where it previously also accepted text file-like objects, like io.StringIO.
|
||||
> * The DocumentConverter class interface has changed to read from file-like streams rather than file paths. *No temporary files are created anymore*. If you are the maintainer of a plugin, or custom DocumentConverter, you likely need to update your code. Otherwise, if only using the MarkItDown class or CLI (as in these examples), you should not need to change anything.
|
||||
|
||||
@@ -132,6 +132,38 @@ markitdown --use-plugins path-to-file.pdf
|
||||
|
||||
To find available plugins, search GitHub for the hashtag `#markitdown-plugin`. To develop a plugin, see `packages/markitdown-sample-plugin`.
|
||||
|
||||
#### markitdown-ocr Plugin
|
||||
|
||||
The `markitdown-ocr` plugin adds OCR support to PDF, DOCX, PPTX, and XLSX converters, extracting text from embedded images using LLM Vision — the same `llm_client` / `llm_model` pattern that MarkItDown already uses for image descriptions. No new ML libraries or binary dependencies required.
|
||||
|
||||
**Installation:**
|
||||
|
||||
```bash
|
||||
pip install markitdown-ocr
|
||||
pip install openai # or any OpenAI-compatible client
|
||||
```
|
||||
|
||||
**Usage:**
|
||||
|
||||
Pass the same `llm_client` and `llm_model` you would use for image descriptions:
|
||||
|
||||
```python
|
||||
from markitdown import MarkItDown
|
||||
from openai import OpenAI
|
||||
|
||||
md = MarkItDown(
|
||||
enable_plugins=True,
|
||||
llm_client=OpenAI(),
|
||||
llm_model="gpt-4o",
|
||||
)
|
||||
result = md.convert("document_with_images.pdf")
|
||||
print(result.text_content)
|
||||
```
|
||||
|
||||
If no `llm_client` is provided the plugin still loads, but OCR is silently skipped and the standard built-in converter is used instead.
|
||||
|
||||
See [`packages/markitdown-ocr/README.md`](packages/markitdown-ocr/README.md) for detailed documentation.
|
||||
|
||||
### Azure Document Intelligence
|
||||
|
||||
To use Microsoft Document Intelligence for conversion:
|
||||
|
||||
@@ -0,0 +1,21 @@
|
||||
MIT License
|
||||
|
||||
Copyright (c) Microsoft Corporation.
|
||||
|
||||
Permission is hereby granted, free of charge, to any person obtaining a copy
|
||||
of this software and associated documentation files (the "Software"), to deal
|
||||
in the Software without restriction, including without limitation the rights
|
||||
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||||
copies of the Software, and to permit persons to whom the Software is
|
||||
furnished to do so, subject to the following conditions:
|
||||
|
||||
The above copyright notice and this permission notice shall be included in all
|
||||
copies or substantial portions of the Software.
|
||||
|
||||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
||||
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||||
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||||
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||||
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||||
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
||||
SOFTWARE
|
||||
@@ -0,0 +1,200 @@
|
||||
# MarkItDown OCR Plugin
|
||||
|
||||
LLM Vision plugin for MarkItDown that extracts text from images embedded in PDF, DOCX, PPTX, and XLSX files.
|
||||
|
||||
Uses the same `llm_client` / `llm_model` pattern that MarkItDown already supports for image descriptions — no new ML libraries or binary dependencies required.
|
||||
|
||||
## Features
|
||||
|
||||
- **Enhanced PDF Converter**: Extracts text from images within PDFs, with full-page OCR fallback for scanned documents
|
||||
- **Enhanced DOCX Converter**: OCR for images in Word documents
|
||||
- **Enhanced PPTX Converter**: OCR for images in PowerPoint presentations
|
||||
- **Enhanced XLSX Converter**: OCR for images in Excel spreadsheets
|
||||
- **Context Preservation**: Maintains document structure and flow when inserting extracted text
|
||||
|
||||
## Installation
|
||||
|
||||
```bash
|
||||
pip install markitdown-ocr
|
||||
```
|
||||
|
||||
The plugin uses whatever OpenAI-compatible client you already have. Install one if you don't have it yet:
|
||||
|
||||
```bash
|
||||
pip install openai
|
||||
```
|
||||
|
||||
## Usage
|
||||
|
||||
### Command Line
|
||||
|
||||
```bash
|
||||
markitdown document.pdf --use-plugins --llm-client openai --llm-model gpt-4o
|
||||
```
|
||||
|
||||
### Python API
|
||||
|
||||
Pass `llm_client` and `llm_model` to `MarkItDown()` exactly as you would for image descriptions:
|
||||
|
||||
```python
|
||||
from markitdown import MarkItDown
|
||||
from openai import OpenAI
|
||||
|
||||
md = MarkItDown(
|
||||
enable_plugins=True,
|
||||
llm_client=OpenAI(),
|
||||
llm_model="gpt-4o",
|
||||
)
|
||||
|
||||
result = md.convert("document_with_images.pdf")
|
||||
print(result.text_content)
|
||||
```
|
||||
|
||||
If no `llm_client` is provided the plugin still loads, but OCR is silently skipped — falling back to the standard built-in converter.
|
||||
|
||||
### Custom Prompt
|
||||
|
||||
Override the default extraction prompt for specialized documents:
|
||||
|
||||
```python
|
||||
md = MarkItDown(
|
||||
enable_plugins=True,
|
||||
llm_client=OpenAI(),
|
||||
llm_model="gpt-4o",
|
||||
llm_prompt="Extract all text from this image, preserving table structure.",
|
||||
)
|
||||
```
|
||||
|
||||
### Any OpenAI-Compatible Client
|
||||
|
||||
Works with any client that follows the OpenAI API:
|
||||
|
||||
```python
|
||||
from openai import AzureOpenAI
|
||||
|
||||
md = MarkItDown(
|
||||
enable_plugins=True,
|
||||
llm_client=AzureOpenAI(
|
||||
api_key="...",
|
||||
azure_endpoint="https://your-resource.openai.azure.com/",
|
||||
api_version="2024-02-01",
|
||||
),
|
||||
llm_model="gpt-4o",
|
||||
)
|
||||
```
|
||||
|
||||
## How It Works
|
||||
|
||||
When `MarkItDown(enable_plugins=True, llm_client=..., llm_model=...)` is called:
|
||||
|
||||
1. MarkItDown discovers the plugin via the `markitdown.plugin` entry point group
|
||||
2. It calls `register_converters()`, forwarding all kwargs including `llm_client` and `llm_model`
|
||||
3. The plugin creates an `LLMVisionOCRService` from those kwargs
|
||||
4. Four OCR-enhanced converters are registered at **priority -1.0** — before the built-in converters at priority 0.0
|
||||
|
||||
When a file is converted:
|
||||
|
||||
1. The OCR converter accepts the file
|
||||
2. It extracts embedded images from the document
|
||||
3. Each image is sent to the LLM with an extraction prompt
|
||||
4. The returned text is inserted inline, preserving document structure
|
||||
5. If the LLM call fails, conversion continues without that image's text
|
||||
|
||||
## Supported File Formats
|
||||
|
||||
### PDF
|
||||
|
||||
- Embedded images are extracted by position (via `page.images` / page XObjects) and OCR'd inline, interleaved with the surrounding text in vertical reading order.
|
||||
- **Scanned PDFs** (pages with no extractable text) are detected automatically: each page is rendered at 300 DPI and sent to the LLM as a full-page image.
|
||||
- **Malformed PDFs** that pdfplumber/pdfminer cannot open (e.g. truncated EOF) are retried with PyMuPDF page rendering, so content is still recovered.
|
||||
|
||||
### DOCX
|
||||
|
||||
- Images are extracted via document part relationships (`doc.part.rels`).
|
||||
- OCR is run before the DOCX→HTML→Markdown pipeline executes: placeholder tokens are injected into the HTML so that the markdown converter does not escape the OCR markers, and the final placeholders are replaced with the formatted `*[Image OCR]...[End OCR]*` blocks after conversion.
|
||||
- Document flow (headings, paragraphs, tables) is fully preserved around the OCR blocks.
|
||||
|
||||
### PPTX
|
||||
|
||||
- Picture shapes, placeholder shapes with images, and images inside groups are all supported.
|
||||
- Shapes are processed in top-to-left reading order per slide.
|
||||
- If an `llm_client` is configured, the LLM is asked for a description first; OCR is used as the fallback when no description is returned.
|
||||
|
||||
### XLSX
|
||||
|
||||
- Images embedded in worksheets (`sheet._images`) are extracted per sheet.
|
||||
- Cell position is calculated from the image anchor coordinates (column/row → Excel letter notation).
|
||||
- Images are listed under a `### Images in this sheet:` section after the sheet's data table — they are not interleaved into the table rows.
|
||||
|
||||
### Output format
|
||||
|
||||
Every extracted OCR block is wrapped as:
|
||||
|
||||
```text
|
||||
*[Image OCR]
|
||||
<extracted text>
|
||||
[End OCR]*
|
||||
```
|
||||
|
||||
## Troubleshooting
|
||||
|
||||
### OCR text missing from output
|
||||
|
||||
The most likely cause is a missing `llm_client` or `llm_model`. Verify:
|
||||
|
||||
```python
|
||||
from openai import OpenAI
|
||||
from markitdown import MarkItDown
|
||||
|
||||
md = MarkItDown(
|
||||
enable_plugins=True,
|
||||
llm_client=OpenAI(), # required
|
||||
llm_model="gpt-4o", # required
|
||||
)
|
||||
```
|
||||
|
||||
### Plugin not loading
|
||||
|
||||
Confirm the plugin is installed and discovered:
|
||||
|
||||
```bash
|
||||
markitdown --list-plugins # should show: ocr
|
||||
```
|
||||
|
||||
### API errors
|
||||
|
||||
The plugin propagates LLM API errors as warnings and continues conversion. Check your API key, quota, and that the chosen model supports vision inputs.
|
||||
|
||||
## Development
|
||||
|
||||
### Running Tests
|
||||
|
||||
```bash
|
||||
cd packages/markitdown-ocr
|
||||
pytest tests/ -v
|
||||
```
|
||||
|
||||
### Building from Source
|
||||
|
||||
```bash
|
||||
git clone https://github.com/microsoft/markitdown.git
|
||||
cd markitdown/packages/markitdown-ocr
|
||||
pip install -e .
|
||||
```
|
||||
|
||||
## Contributing
|
||||
|
||||
Contributions are welcome! See the [MarkItDown repository](https://github.com/microsoft/markitdown) for guidelines.
|
||||
|
||||
## License
|
||||
|
||||
MIT — see [LICENSE](LICENSE).
|
||||
|
||||
## Changelog
|
||||
|
||||
### 0.1.0 (Initial Release)
|
||||
|
||||
- LLM Vision OCR for PDF, DOCX, PPTX, XLSX
|
||||
- Full-page OCR fallback for scanned PDFs
|
||||
- Context-aware inline text insertion
|
||||
- Priority-based converter replacement (no code changes required)
|
||||
@@ -0,0 +1,57 @@
|
||||
[build-system]
|
||||
requires = ["hatchling"]
|
||||
build-backend = "hatchling.build"
|
||||
|
||||
[project]
|
||||
name = "markitdown-ocr"
|
||||
dynamic = ["version"]
|
||||
description = 'OCR plugin for MarkItDown - Extracts text from images in PDF, DOCX, PPTX, and XLSX via LLM Vision'
|
||||
readme = "README.md"
|
||||
requires-python = ">=3.10"
|
||||
license = "MIT"
|
||||
keywords = ["markitdown", "ocr", "pdf", "docx", "xlsx", "pptx", "llm", "vision"]
|
||||
authors = [
|
||||
{ name = "Contributors", email = "noreply@github.com" },
|
||||
]
|
||||
classifiers = [
|
||||
"Development Status :: 4 - Beta",
|
||||
"Programming Language :: Python",
|
||||
"Programming Language :: Python :: 3.10",
|
||||
"Programming Language :: Python :: 3.11",
|
||||
"Programming Language :: Python :: 3.12",
|
||||
"Programming Language :: Python :: 3.13",
|
||||
"Programming Language :: Python :: Implementation :: CPython",
|
||||
]
|
||||
|
||||
# Core dependencies — matches the file-format libraries markitdown already uses
|
||||
dependencies = [
|
||||
"markitdown>=0.1.0",
|
||||
"pdfminer.six>=20251230",
|
||||
"pdfplumber>=0.11.9",
|
||||
"PyMuPDF>=1.24.0",
|
||||
"mammoth~=1.11.0",
|
||||
"python-docx",
|
||||
"python-pptx",
|
||||
"pandas",
|
||||
"openpyxl",
|
||||
"Pillow>=9.0.0",
|
||||
]
|
||||
|
||||
# llm_client is passed in by the user (same as for markitdown image descriptions);
|
||||
# install openai or any OpenAI-compatible SDK separately.
|
||||
[project.optional-dependencies]
|
||||
llm = [
|
||||
"openai>=1.0.0",
|
||||
]
|
||||
|
||||
[project.urls]
|
||||
Documentation = "https://github.com/microsoft/markitdown#readme"
|
||||
Issues = "https://github.com/microsoft/markitdown/issues"
|
||||
Source = "https://github.com/microsoft/markitdown"
|
||||
|
||||
[tool.hatch.version]
|
||||
path = "src/markitdown_ocr/__about__.py"
|
||||
|
||||
# CRITICAL: Plugin entry point - MarkItDown will discover this plugin through this entry point
|
||||
[project.entry-points."markitdown.plugin"]
|
||||
ocr = "markitdown_ocr"
|
||||
@@ -0,0 +1,4 @@
|
||||
# SPDX-FileCopyrightText: 2025-present Contributors
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
__version__ = "0.1.0"
|
||||
@@ -0,0 +1,31 @@
|
||||
# SPDX-FileCopyrightText: 2025-present Contributors
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
"""
|
||||
markitdown-ocr: OCR plugin for MarkItDown
|
||||
|
||||
Adds LLM Vision-based text extraction from images embedded in PDF, DOCX, PPTX, and XLSX files.
|
||||
"""
|
||||
|
||||
from ._plugin import __plugin_interface_version__, register_converters
|
||||
from .__about__ import __version__
|
||||
from ._ocr_service import (
|
||||
OCRResult,
|
||||
LLMVisionOCRService,
|
||||
)
|
||||
from ._pdf_converter_with_ocr import PdfConverterWithOCR
|
||||
from ._docx_converter_with_ocr import DocxConverterWithOCR
|
||||
from ._pptx_converter_with_ocr import PptxConverterWithOCR
|
||||
from ._xlsx_converter_with_ocr import XlsxConverterWithOCR
|
||||
|
||||
__all__ = [
|
||||
"__version__",
|
||||
"__plugin_interface_version__",
|
||||
"register_converters",
|
||||
"OCRResult",
|
||||
"LLMVisionOCRService",
|
||||
"PdfConverterWithOCR",
|
||||
"DocxConverterWithOCR",
|
||||
"PptxConverterWithOCR",
|
||||
"XlsxConverterWithOCR",
|
||||
]
|
||||
@@ -0,0 +1,189 @@
|
||||
"""
|
||||
Enhanced DOCX Converter with OCR support for embedded images.
|
||||
Extracts images from Word documents and performs OCR while maintaining context.
|
||||
"""
|
||||
|
||||
import io
|
||||
import re
|
||||
import sys
|
||||
from typing import Any, BinaryIO, Optional
|
||||
|
||||
from markitdown.converters import HtmlConverter
|
||||
from markitdown.converter_utils.docx.pre_process import pre_process_docx
|
||||
from markitdown import DocumentConverterResult, StreamInfo
|
||||
from markitdown._exceptions import (
|
||||
MissingDependencyException,
|
||||
MISSING_DEPENDENCY_MESSAGE,
|
||||
)
|
||||
from ._ocr_service import LLMVisionOCRService
|
||||
|
||||
# Try loading dependencies
|
||||
_dependency_exc_info = None
|
||||
try:
|
||||
import mammoth
|
||||
from docx import Document
|
||||
except ImportError:
|
||||
_dependency_exc_info = sys.exc_info()
|
||||
|
||||
# Placeholder injected into HTML so that mammoth never sees the OCR markers.
|
||||
# Must be a single token with no special markdown characters.
|
||||
_PLACEHOLDER = "MARKITDOWNOCRBLOCK{}"
|
||||
|
||||
|
||||
class DocxConverterWithOCR(HtmlConverter):
|
||||
"""
|
||||
Enhanced DOCX Converter with OCR support for embedded images.
|
||||
Maintains document flow while extracting text from images inline.
|
||||
"""
|
||||
|
||||
def __init__(self, ocr_service: Optional[LLMVisionOCRService] = None):
|
||||
super().__init__()
|
||||
self._html_converter = HtmlConverter()
|
||||
self.ocr_service = ocr_service
|
||||
|
||||
def accepts(
|
||||
self,
|
||||
file_stream: BinaryIO,
|
||||
stream_info: StreamInfo,
|
||||
**kwargs: Any,
|
||||
) -> bool:
|
||||
mimetype = (stream_info.mimetype or "").lower()
|
||||
extension = (stream_info.extension or "").lower()
|
||||
|
||||
if extension == ".docx":
|
||||
return True
|
||||
|
||||
if mimetype.startswith(
|
||||
"application/vnd.openxmlformats-officedocument.wordprocessingml"
|
||||
):
|
||||
return True
|
||||
|
||||
return False
|
||||
|
||||
def convert(
|
||||
self,
|
||||
file_stream: BinaryIO,
|
||||
stream_info: StreamInfo,
|
||||
**kwargs: Any,
|
||||
) -> DocumentConverterResult:
|
||||
if _dependency_exc_info is not None:
|
||||
raise MissingDependencyException(
|
||||
MISSING_DEPENDENCY_MESSAGE.format(
|
||||
converter=type(self).__name__,
|
||||
extension=".docx",
|
||||
feature="docx",
|
||||
)
|
||||
) from _dependency_exc_info[1].with_traceback(
|
||||
_dependency_exc_info[2]
|
||||
) # type: ignore[union-attr]
|
||||
|
||||
# Get OCR service if available (from kwargs or instance)
|
||||
ocr_service: Optional[LLMVisionOCRService] = (
|
||||
kwargs.get("ocr_service") or self.ocr_service
|
||||
)
|
||||
|
||||
if ocr_service:
|
||||
# 1. Extract and OCR images — returns raw text per image
|
||||
file_stream.seek(0)
|
||||
image_ocr_map = self._extract_and_ocr_images(file_stream, ocr_service)
|
||||
|
||||
# 2. Convert DOCX → HTML via mammoth
|
||||
file_stream.seek(0)
|
||||
pre_process_stream = pre_process_docx(file_stream)
|
||||
html_result = mammoth.convert_to_html(
|
||||
pre_process_stream, style_map=kwargs.get("style_map")
|
||||
).value
|
||||
|
||||
# 3. Replace <img> tags with plain placeholder tokens so that
|
||||
# mammoth's HTML→markdown step never escapes our OCR markers.
|
||||
html_with_placeholders, ocr_texts = self._inject_placeholders(
|
||||
html_result, image_ocr_map
|
||||
)
|
||||
|
||||
# 4. Convert HTML → markdown
|
||||
md_result = self._html_converter.convert_string(
|
||||
html_with_placeholders, **kwargs
|
||||
)
|
||||
md = md_result.markdown
|
||||
|
||||
# 5. Swap placeholders for the actual OCR blocks (post-conversion
|
||||
# so * and _ are never escaped by the markdown converter).
|
||||
for i, raw_text in enumerate(ocr_texts):
|
||||
placeholder = _PLACEHOLDER.format(i)
|
||||
ocr_block = f"*[Image OCR]\n{raw_text}\n[End OCR]*"
|
||||
md = md.replace(placeholder, ocr_block)
|
||||
|
||||
return DocumentConverterResult(markdown=md)
|
||||
else:
|
||||
# Standard conversion without OCR
|
||||
style_map = kwargs.get("style_map", None)
|
||||
pre_process_stream = pre_process_docx(file_stream)
|
||||
return self._html_converter.convert_string(
|
||||
mammoth.convert_to_html(pre_process_stream, style_map=style_map).value,
|
||||
**kwargs,
|
||||
)
|
||||
|
||||
def _extract_and_ocr_images(
|
||||
self, file_stream: BinaryIO, ocr_service: LLMVisionOCRService
|
||||
) -> dict[str, str]:
|
||||
"""
|
||||
Extract images from DOCX and OCR them.
|
||||
|
||||
Returns:
|
||||
Dict mapping image relationship IDs to raw OCR text (no markers).
|
||||
"""
|
||||
ocr_map = {}
|
||||
|
||||
try:
|
||||
file_stream.seek(0)
|
||||
doc = Document(file_stream)
|
||||
|
||||
for rel in doc.part.rels.values():
|
||||
if "image" in rel.target_ref.lower():
|
||||
try:
|
||||
image_bytes = rel.target_part.blob
|
||||
image_stream = io.BytesIO(image_bytes)
|
||||
ocr_result = ocr_service.extract_text(image_stream)
|
||||
|
||||
if ocr_result.text.strip():
|
||||
# Store raw text only — markers added later
|
||||
ocr_map[rel.rId] = ocr_result.text.strip()
|
||||
|
||||
except Exception:
|
||||
continue
|
||||
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
return ocr_map
|
||||
|
||||
def _inject_placeholders(
|
||||
self, html: str, ocr_map: dict[str, str]
|
||||
) -> tuple[str, list[str]]:
|
||||
"""
|
||||
Replace <img> tags with numbered placeholder tokens.
|
||||
|
||||
Returns:
|
||||
(html_with_placeholders, ordered list of raw OCR texts)
|
||||
"""
|
||||
if not ocr_map:
|
||||
return html, []
|
||||
|
||||
ocr_texts = list(ocr_map.values())
|
||||
used: list[int] = []
|
||||
|
||||
def replace_img(match: re.Match) -> str: # type: ignore[type-arg]
|
||||
for i in range(len(ocr_texts)):
|
||||
if i not in used:
|
||||
used.append(i)
|
||||
return f"<p>{_PLACEHOLDER.format(i)}</p>"
|
||||
return "" # remove image if all OCR texts already used
|
||||
|
||||
result = re.sub(r"<img[^>]*>", replace_img, html)
|
||||
|
||||
# Any OCR texts that had no matching <img> tag go at the end
|
||||
for i in range(len(ocr_texts)):
|
||||
if i not in used:
|
||||
result += f"<p>{_PLACEHOLDER.format(i)}</p>"
|
||||
|
||||
return result, ocr_texts
|
||||
@@ -0,0 +1,110 @@
|
||||
"""
|
||||
OCR Service Layer for MarkItDown
|
||||
Provides LLM Vision-based image text extraction.
|
||||
"""
|
||||
|
||||
import base64
|
||||
from typing import Any, BinaryIO
|
||||
from dataclasses import dataclass
|
||||
|
||||
from markitdown import StreamInfo
|
||||
|
||||
|
||||
@dataclass
|
||||
class OCRResult:
|
||||
"""Result from OCR extraction."""
|
||||
|
||||
text: str
|
||||
confidence: float | None = None
|
||||
backend_used: str | None = None
|
||||
error: str | None = None
|
||||
|
||||
|
||||
class LLMVisionOCRService:
|
||||
"""OCR service using LLM vision models (OpenAI-compatible)."""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
client: Any,
|
||||
model: str,
|
||||
default_prompt: str | None = None,
|
||||
) -> None:
|
||||
"""
|
||||
Initialize LLM Vision OCR service.
|
||||
|
||||
Args:
|
||||
client: OpenAI-compatible client
|
||||
model: Model name (e.g., 'gpt-4o', 'gemini-2.0-flash')
|
||||
default_prompt: Default prompt for OCR extraction
|
||||
"""
|
||||
self.client = client
|
||||
self.model = model
|
||||
self.default_prompt = default_prompt or (
|
||||
"Extract all text from this image. "
|
||||
"Return ONLY the extracted text, maintaining the original "
|
||||
"layout and order. Do not add any commentary or description."
|
||||
)
|
||||
|
||||
def extract_text(
|
||||
self,
|
||||
image_stream: BinaryIO,
|
||||
prompt: str | None = None,
|
||||
stream_info: StreamInfo | None = None,
|
||||
**kwargs: Any,
|
||||
) -> OCRResult:
|
||||
"""Extract text using LLM vision."""
|
||||
if self.client is None:
|
||||
return OCRResult(
|
||||
text="",
|
||||
backend_used="llm_vision",
|
||||
error="LLM client not configured",
|
||||
)
|
||||
|
||||
try:
|
||||
image_stream.seek(0)
|
||||
|
||||
content_type: str | None = None
|
||||
if stream_info:
|
||||
content_type = stream_info.mimetype
|
||||
|
||||
if not content_type:
|
||||
try:
|
||||
from PIL import Image
|
||||
|
||||
image_stream.seek(0)
|
||||
img = Image.open(image_stream)
|
||||
fmt = img.format.lower() if img.format else "png"
|
||||
content_type = f"image/{fmt}"
|
||||
except Exception:
|
||||
content_type = "image/png"
|
||||
|
||||
image_stream.seek(0)
|
||||
base64_image = base64.b64encode(image_stream.read()).decode("utf-8")
|
||||
data_uri = f"data:{content_type};base64,{base64_image}"
|
||||
|
||||
actual_prompt = prompt or self.default_prompt
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": actual_prompt},
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": data_uri},
|
||||
},
|
||||
],
|
||||
}
|
||||
],
|
||||
)
|
||||
|
||||
text = response.choices[0].message.content
|
||||
return OCRResult(
|
||||
text=text.strip() if text else "",
|
||||
backend_used="llm_vision",
|
||||
)
|
||||
except Exception as e:
|
||||
return OCRResult(text="", backend_used="llm_vision", error=str(e))
|
||||
finally:
|
||||
image_stream.seek(0)
|
||||
@@ -0,0 +1,422 @@
|
||||
"""
|
||||
Enhanced PDF Converter with OCR support for embedded images.
|
||||
Extracts images from PDFs and performs OCR while maintaining document context.
|
||||
"""
|
||||
|
||||
import io
|
||||
import sys
|
||||
from typing import Any, BinaryIO, Optional
|
||||
|
||||
from markitdown import DocumentConverter, DocumentConverterResult, StreamInfo
|
||||
from markitdown._exceptions import (
|
||||
MissingDependencyException,
|
||||
MISSING_DEPENDENCY_MESSAGE,
|
||||
)
|
||||
from ._ocr_service import LLMVisionOCRService
|
||||
|
||||
# Import dependencies
|
||||
_dependency_exc_info = None
|
||||
try:
|
||||
import pdfminer
|
||||
import pdfminer.high_level
|
||||
import pdfplumber
|
||||
from PIL import Image
|
||||
except ImportError:
|
||||
_dependency_exc_info = sys.exc_info()
|
||||
|
||||
|
||||
def _extract_images_from_page(page: Any) -> list[dict]:
|
||||
"""
|
||||
Extract images from a PDF page by rendering page regions.
|
||||
|
||||
Returns:
|
||||
List of dicts with 'stream', 'bbox', 'name', 'y_pos' keys
|
||||
"""
|
||||
images_info = []
|
||||
|
||||
try:
|
||||
# Try multiple methods to detect images
|
||||
images = []
|
||||
|
||||
# Method 1: Use page.images (standard approach)
|
||||
if hasattr(page, "images") and page.images:
|
||||
images = page.images
|
||||
|
||||
# Method 2: If no images found, try underlying PDF objects
|
||||
if not images and hasattr(page, "objects") and "image" in page.objects:
|
||||
images = page.objects.get("image", [])
|
||||
|
||||
# Method 3: Try filtering all objects for image types
|
||||
if not images and hasattr(page, "objects"):
|
||||
all_objs = page.objects
|
||||
for obj_type in all_objs.keys():
|
||||
if "image" in obj_type.lower() or "xobject" in obj_type.lower():
|
||||
potential_imgs = all_objs.get(obj_type, [])
|
||||
if potential_imgs:
|
||||
images = potential_imgs
|
||||
break
|
||||
|
||||
for i, img_dict in enumerate(images):
|
||||
try:
|
||||
# Try to get the actual image stream from the PDF
|
||||
img_stream = None
|
||||
y_pos = 0
|
||||
|
||||
# Method A: If img_dict has 'stream' key, use it directly
|
||||
if "stream" in img_dict and hasattr(img_dict["stream"], "get_data"):
|
||||
try:
|
||||
img_bytes = img_dict["stream"].get_data()
|
||||
|
||||
# Try to open as PIL Image to validate/decode
|
||||
pil_img = Image.open(io.BytesIO(img_bytes))
|
||||
|
||||
# Convert to RGB if needed (handle CMYK, etc.)
|
||||
if pil_img.mode not in ("RGB", "L"):
|
||||
pil_img = pil_img.convert("RGB")
|
||||
|
||||
# Save to stream as PNG
|
||||
img_stream = io.BytesIO()
|
||||
pil_img.save(img_stream, format="PNG")
|
||||
img_stream.seek(0)
|
||||
|
||||
y_pos = img_dict.get("top", 0)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# Method B: Fallback to rendering page region
|
||||
if img_stream is None:
|
||||
x0 = img_dict.get("x0", 0)
|
||||
y0 = img_dict.get("top", 0)
|
||||
x1 = img_dict.get("x1", 0)
|
||||
y1 = img_dict.get("bottom", 0)
|
||||
y_pos = y0
|
||||
|
||||
# Check if dimensions are valid
|
||||
if x1 <= x0 or y1 <= y0:
|
||||
continue
|
||||
|
||||
# Use pdfplumber's within_bbox to crop, then render
|
||||
# This preserves coordinate system correctly
|
||||
bbox = (x0, y0, x1, y1)
|
||||
cropped_page = page.within_bbox(bbox)
|
||||
|
||||
# Render at 150 DPI (balance between quality and size)
|
||||
page_img = cropped_page.to_image(resolution=150)
|
||||
|
||||
# Save to stream
|
||||
img_stream = io.BytesIO()
|
||||
page_img.original.save(img_stream, format="PNG")
|
||||
img_stream.seek(0)
|
||||
|
||||
if img_stream:
|
||||
images_info.append(
|
||||
{
|
||||
"stream": img_stream,
|
||||
"name": f"page_{page.page_number}_img_{i}",
|
||||
"y_pos": y_pos,
|
||||
}
|
||||
)
|
||||
|
||||
except Exception:
|
||||
continue
|
||||
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
return images_info
|
||||
|
||||
|
||||
class PdfConverterWithOCR(DocumentConverter):
|
||||
"""
|
||||
Enhanced PDF Converter with OCR support for embedded images.
|
||||
Maintains document structure while extracting text from images inline.
|
||||
"""
|
||||
|
||||
def __init__(self, ocr_service: Optional[LLMVisionOCRService] = None):
|
||||
super().__init__()
|
||||
self.ocr_service = ocr_service
|
||||
|
||||
def accepts(
|
||||
self,
|
||||
file_stream: BinaryIO,
|
||||
stream_info: StreamInfo,
|
||||
**kwargs: Any,
|
||||
) -> bool:
|
||||
mimetype = (stream_info.mimetype or "").lower()
|
||||
extension = (stream_info.extension or "").lower()
|
||||
|
||||
if extension == ".pdf":
|
||||
return True
|
||||
|
||||
if mimetype.startswith("application/pdf") or mimetype.startswith(
|
||||
"application/x-pdf"
|
||||
):
|
||||
return True
|
||||
|
||||
return False
|
||||
|
||||
def convert(
|
||||
self,
|
||||
file_stream: BinaryIO,
|
||||
stream_info: StreamInfo,
|
||||
**kwargs: Any,
|
||||
) -> DocumentConverterResult:
|
||||
if _dependency_exc_info is not None:
|
||||
raise MissingDependencyException(
|
||||
MISSING_DEPENDENCY_MESSAGE.format(
|
||||
converter=type(self).__name__,
|
||||
extension=".pdf",
|
||||
feature="pdf",
|
||||
)
|
||||
) from _dependency_exc_info[1].with_traceback(
|
||||
_dependency_exc_info[2]
|
||||
) # type: ignore[union-attr]
|
||||
|
||||
# Get OCR service if available (from kwargs or instance)
|
||||
ocr_service: LLMVisionOCRService | None = (
|
||||
kwargs.get("ocr_service") or self.ocr_service
|
||||
)
|
||||
|
||||
# Read PDF into BytesIO
|
||||
file_stream.seek(0)
|
||||
pdf_bytes = io.BytesIO(file_stream.read())
|
||||
|
||||
markdown_content = []
|
||||
|
||||
try:
|
||||
with pdfplumber.open(pdf_bytes) as pdf:
|
||||
for page_num, page in enumerate(pdf.pages, 1):
|
||||
markdown_content.append(f"\n## Page {page_num}\n")
|
||||
|
||||
# If OCR is enabled, interleave text and images by position
|
||||
if ocr_service:
|
||||
images_on_page = self._extract_page_images(pdf_bytes, page_num)
|
||||
|
||||
if images_on_page:
|
||||
# Extract text lines with Y positions
|
||||
chars = page.chars
|
||||
if chars:
|
||||
# Group chars into lines based on Y position
|
||||
lines_with_y = []
|
||||
current_line = []
|
||||
current_y = None
|
||||
|
||||
for char in sorted(
|
||||
chars, key=lambda c: (c["top"], c["x0"])
|
||||
):
|
||||
y = char["top"]
|
||||
if current_y is None:
|
||||
current_y = y
|
||||
elif abs(y - current_y) > 2: # New line threshold
|
||||
if current_line:
|
||||
text = "".join(
|
||||
[c["text"] for c in current_line]
|
||||
)
|
||||
lines_with_y.append(
|
||||
{"y": current_y, "text": text.strip()}
|
||||
)
|
||||
current_line = []
|
||||
current_y = y
|
||||
current_line.append(char)
|
||||
|
||||
# Add last line
|
||||
if current_line:
|
||||
text = "".join([c["text"] for c in current_line])
|
||||
lines_with_y.append(
|
||||
{"y": current_y, "text": text.strip()}
|
||||
)
|
||||
else:
|
||||
# Fallback: use simple text extraction
|
||||
text_content = page.extract_text() or ""
|
||||
lines_with_y = [
|
||||
{"y": i * 10, "text": line}
|
||||
for i, line in enumerate(text_content.split("\n"))
|
||||
]
|
||||
|
||||
# OCR all images
|
||||
image_data = []
|
||||
for img_info in images_on_page:
|
||||
ocr_result = ocr_service.extract_text(
|
||||
img_info["stream"]
|
||||
)
|
||||
if ocr_result.text.strip():
|
||||
image_data.append(
|
||||
{
|
||||
"y_pos": img_info["y_pos"],
|
||||
"name": img_info["name"],
|
||||
"ocr_text": ocr_result.text,
|
||||
"backend": ocr_result.backend_used,
|
||||
"type": "image",
|
||||
}
|
||||
)
|
||||
|
||||
# Add text items
|
||||
content_items = [
|
||||
{
|
||||
"y_pos": item["y"],
|
||||
"text": item["text"],
|
||||
"type": "text",
|
||||
}
|
||||
for item in lines_with_y
|
||||
if item["text"]
|
||||
]
|
||||
content_items.extend(image_data)
|
||||
|
||||
# Sort all items by Y position (top to bottom)
|
||||
content_items.sort(key=lambda x: x["y_pos"])
|
||||
|
||||
# Build markdown by interleaving text and images
|
||||
for item in content_items:
|
||||
if item["type"] == "text":
|
||||
markdown_content.append(item["text"])
|
||||
else: # image
|
||||
ocr_text = item["ocr_text"]
|
||||
img_marker = (
|
||||
f"\n\n*[Image OCR]\n{ocr_text}\n[End OCR]*\n"
|
||||
)
|
||||
markdown_content.append(img_marker)
|
||||
else:
|
||||
# No images detected - just extract regular text
|
||||
text_content = page.extract_text() or ""
|
||||
if text_content.strip():
|
||||
markdown_content.append(text_content.strip())
|
||||
else:
|
||||
# No OCR, just extract text
|
||||
text_content = page.extract_text() or ""
|
||||
if text_content.strip():
|
||||
markdown_content.append(text_content.strip())
|
||||
|
||||
# Build final markdown
|
||||
markdown = "\n\n".join(markdown_content).strip()
|
||||
|
||||
# Fallback to pdfminer if empty
|
||||
if not markdown:
|
||||
pdf_bytes.seek(0)
|
||||
markdown = pdfminer.high_level.extract_text(pdf_bytes)
|
||||
|
||||
except Exception:
|
||||
# Fallback to pdfminer
|
||||
try:
|
||||
pdf_bytes.seek(0)
|
||||
markdown = pdfminer.high_level.extract_text(pdf_bytes)
|
||||
except Exception:
|
||||
markdown = ""
|
||||
|
||||
# Final fallback: If still empty/whitespace and OCR is available,
|
||||
# treat as scanned PDF and OCR full pages
|
||||
if ocr_service and (not markdown or not markdown.strip()):
|
||||
pdf_bytes.seek(0)
|
||||
markdown = self._ocr_full_pages(pdf_bytes, ocr_service)
|
||||
|
||||
return DocumentConverterResult(markdown=markdown)
|
||||
|
||||
def _extract_page_images(self, pdf_bytes: io.BytesIO, page_num: int) -> list[dict]:
|
||||
"""
|
||||
Extract images from a PDF page using pdfplumber.
|
||||
|
||||
Args:
|
||||
pdf_bytes: PDF file as BytesIO
|
||||
page_num: Page number (1-indexed)
|
||||
|
||||
Returns:
|
||||
List of image info dicts with 'stream', 'bbox', 'name', 'y_pos'
|
||||
"""
|
||||
images = []
|
||||
|
||||
try:
|
||||
pdf_bytes.seek(0)
|
||||
with pdfplumber.open(pdf_bytes) as pdf:
|
||||
if page_num <= len(pdf.pages):
|
||||
page = pdf.pages[page_num - 1] # 0-indexed
|
||||
images = _extract_images_from_page(page)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# Sort by vertical position (top to bottom)
|
||||
images.sort(key=lambda x: x["y_pos"])
|
||||
|
||||
return images
|
||||
|
||||
def _ocr_full_pages(
|
||||
self, pdf_bytes: io.BytesIO, ocr_service: LLMVisionOCRService
|
||||
) -> str:
|
||||
"""
|
||||
Fallback for scanned PDFs: Convert entire pages to images and OCR them.
|
||||
Used when text extraction returns empty/whitespace results.
|
||||
|
||||
Args:
|
||||
pdf_bytes: PDF file as BytesIO
|
||||
ocr_service: OCR service to use
|
||||
|
||||
Returns:
|
||||
Markdown text extracted from OCR of full pages
|
||||
"""
|
||||
markdown_parts = []
|
||||
|
||||
try:
|
||||
pdf_bytes.seek(0)
|
||||
with pdfplumber.open(pdf_bytes) as pdf:
|
||||
for page_num, page in enumerate(pdf.pages, 1):
|
||||
try:
|
||||
markdown_parts.append(f"\n## Page {page_num}\n")
|
||||
|
||||
# Render page to image
|
||||
page_img = page.to_image(resolution=300)
|
||||
img_stream = io.BytesIO()
|
||||
page_img.original.save(img_stream, format="PNG")
|
||||
img_stream.seek(0)
|
||||
|
||||
# Run OCR
|
||||
ocr_result = ocr_service.extract_text(img_stream)
|
||||
|
||||
if ocr_result.text.strip():
|
||||
text = ocr_result.text.strip()
|
||||
markdown_parts.append(f"*[Image OCR]\n{text}\n[End OCR]*")
|
||||
else:
|
||||
markdown_parts.append(
|
||||
"*[No text could be extracted from this page]*"
|
||||
)
|
||||
|
||||
except Exception as e:
|
||||
markdown_parts.append(
|
||||
f"*[Error processing page {page_num}: {str(e)}]*"
|
||||
)
|
||||
continue
|
||||
|
||||
except Exception:
|
||||
# pdfplumber failed (e.g. malformed EOF) — try PyMuPDF for rendering
|
||||
markdown_parts = []
|
||||
try:
|
||||
import fitz # PyMuPDF
|
||||
|
||||
pdf_bytes.seek(0)
|
||||
doc = fitz.open(stream=pdf_bytes.read(), filetype="pdf")
|
||||
for page_num in range(1, doc.page_count + 1):
|
||||
try:
|
||||
markdown_parts.append(f"\n## Page {page_num}\n")
|
||||
page = doc[page_num - 1]
|
||||
mat = fitz.Matrix(300 / 72, 300 / 72) # 300 DPI
|
||||
pix = page.get_pixmap(matrix=mat)
|
||||
img_stream = io.BytesIO(pix.tobytes("png"))
|
||||
img_stream.seek(0)
|
||||
|
||||
ocr_result = ocr_service.extract_text(img_stream)
|
||||
|
||||
if ocr_result.text.strip():
|
||||
text = ocr_result.text.strip()
|
||||
markdown_parts.append(f"*[Image OCR]\n{text}\n[End OCR]*")
|
||||
else:
|
||||
markdown_parts.append(
|
||||
"*[No text could be extracted from this page]*"
|
||||
)
|
||||
|
||||
except Exception as e:
|
||||
markdown_parts.append(
|
||||
f"*[Error processing page {page_num}: {str(e)}]*"
|
||||
)
|
||||
continue
|
||||
doc.close()
|
||||
except Exception:
|
||||
return "*[Error: Could not process scanned PDF]*"
|
||||
|
||||
return "\n\n".join(markdown_parts).strip()
|
||||
@@ -0,0 +1,68 @@
|
||||
"""
|
||||
Plugin registration for markitdown-ocr.
|
||||
Registers OCR-enhanced converters with priority-based replacement strategy.
|
||||
"""
|
||||
|
||||
from typing import Any
|
||||
from markitdown import MarkItDown
|
||||
|
||||
from ._ocr_service import LLMVisionOCRService
|
||||
from ._pdf_converter_with_ocr import PdfConverterWithOCR
|
||||
from ._docx_converter_with_ocr import DocxConverterWithOCR
|
||||
from ._pptx_converter_with_ocr import PptxConverterWithOCR
|
||||
from ._xlsx_converter_with_ocr import XlsxConverterWithOCR
|
||||
|
||||
|
||||
__plugin_interface_version__ = 1
|
||||
|
||||
|
||||
def register_converters(markitdown: MarkItDown, **kwargs: Any) -> None:
|
||||
"""
|
||||
Register OCR-enhanced converters with MarkItDown.
|
||||
|
||||
This plugin provides OCR support for PDF, DOCX, PPTX, and XLSX files.
|
||||
The converters are registered with priority -1.0 to run BEFORE built-in
|
||||
converters (which have priority 0.0), effectively replacing them when
|
||||
the plugin is enabled.
|
||||
|
||||
Args:
|
||||
markitdown: MarkItDown instance to register converters with
|
||||
**kwargs: Additional keyword arguments that may include:
|
||||
- llm_client: OpenAI-compatible client for LLM-based OCR (required for OCR to work)
|
||||
- llm_model: Model name (e.g., 'gpt-4o')
|
||||
- llm_prompt: Custom prompt for text extraction
|
||||
"""
|
||||
# Create OCR service — reads the same llm_client/llm_model kwargs
|
||||
# that MarkItDown itself already accepts for image descriptions
|
||||
llm_client = kwargs.get("llm_client")
|
||||
llm_model = kwargs.get("llm_model")
|
||||
llm_prompt = kwargs.get("llm_prompt")
|
||||
|
||||
ocr_service: LLMVisionOCRService | None = None
|
||||
if llm_client and llm_model:
|
||||
ocr_service = LLMVisionOCRService(
|
||||
client=llm_client,
|
||||
model=llm_model,
|
||||
default_prompt=llm_prompt,
|
||||
)
|
||||
|
||||
# Register converters with priority -1.0 (before built-ins at 0.0)
|
||||
# This effectively "replaces" the built-in converters when plugin is installed
|
||||
# Pass the OCR service to each converter's constructor
|
||||
PRIORITY_OCR_ENHANCED = -1.0
|
||||
|
||||
markitdown.register_converter(
|
||||
PdfConverterWithOCR(ocr_service=ocr_service), priority=PRIORITY_OCR_ENHANCED
|
||||
)
|
||||
|
||||
markitdown.register_converter(
|
||||
DocxConverterWithOCR(ocr_service=ocr_service), priority=PRIORITY_OCR_ENHANCED
|
||||
)
|
||||
|
||||
markitdown.register_converter(
|
||||
PptxConverterWithOCR(ocr_service=ocr_service), priority=PRIORITY_OCR_ENHANCED
|
||||
)
|
||||
|
||||
markitdown.register_converter(
|
||||
XlsxConverterWithOCR(ocr_service=ocr_service), priority=PRIORITY_OCR_ENHANCED
|
||||
)
|
||||
@@ -0,0 +1,249 @@
|
||||
"""
|
||||
Enhanced PPTX Converter with improved OCR support.
|
||||
Already has LLM-based image description, this enhances it with traditional OCR fallback.
|
||||
"""
|
||||
|
||||
import io
|
||||
import sys
|
||||
from typing import Any, BinaryIO, Optional
|
||||
|
||||
from typing import BinaryIO, Any, Optional
|
||||
|
||||
from markitdown.converters import HtmlConverter
|
||||
from markitdown import DocumentConverter, DocumentConverterResult, StreamInfo
|
||||
from markitdown._exceptions import (
|
||||
MissingDependencyException,
|
||||
MISSING_DEPENDENCY_MESSAGE,
|
||||
)
|
||||
from ._ocr_service import LLMVisionOCRService
|
||||
|
||||
_dependency_exc_info = None
|
||||
try:
|
||||
import pptx
|
||||
except ImportError:
|
||||
_dependency_exc_info = sys.exc_info()
|
||||
|
||||
|
||||
class PptxConverterWithOCR(DocumentConverter):
|
||||
"""Enhanced PPTX Converter with OCR fallback."""
|
||||
|
||||
def __init__(self, ocr_service: Optional[LLMVisionOCRService] = None):
|
||||
super().__init__()
|
||||
self._html_converter = HtmlConverter()
|
||||
self.ocr_service = ocr_service
|
||||
|
||||
def accepts(
|
||||
self,
|
||||
file_stream: BinaryIO,
|
||||
stream_info: StreamInfo,
|
||||
**kwargs: Any,
|
||||
) -> bool:
|
||||
mimetype = (stream_info.mimetype or "").lower()
|
||||
extension = (stream_info.extension or "").lower()
|
||||
|
||||
if extension == ".pptx":
|
||||
return True
|
||||
|
||||
if mimetype.startswith(
|
||||
"application/vnd.openxmlformats-officedocument.presentationml"
|
||||
):
|
||||
return True
|
||||
|
||||
return False
|
||||
|
||||
def convert(
|
||||
self,
|
||||
file_stream: BinaryIO,
|
||||
stream_info: StreamInfo,
|
||||
**kwargs: Any,
|
||||
) -> DocumentConverterResult:
|
||||
if _dependency_exc_info is not None:
|
||||
raise MissingDependencyException(
|
||||
MISSING_DEPENDENCY_MESSAGE.format(
|
||||
converter=type(self).__name__,
|
||||
extension=".pptx",
|
||||
feature="pptx",
|
||||
)
|
||||
) from _dependency_exc_info[1].with_traceback(
|
||||
_dependency_exc_info[2]
|
||||
) # type: ignore[union-attr]
|
||||
|
||||
# Get OCR service (from kwargs or instance)
|
||||
ocr_service: Optional[LLMVisionOCRService] = (
|
||||
kwargs.get("ocr_service") or self.ocr_service
|
||||
)
|
||||
llm_client = kwargs.get("llm_client")
|
||||
|
||||
presentation = pptx.Presentation(file_stream)
|
||||
md_content = ""
|
||||
slide_num = 0
|
||||
|
||||
for slide in presentation.slides:
|
||||
slide_num += 1
|
||||
md_content += f"\\n\\n<!-- Slide number: {slide_num} -->\\n"
|
||||
|
||||
title = slide.shapes.title
|
||||
|
||||
def get_shape_content(shape, **kwargs):
|
||||
nonlocal md_content
|
||||
|
||||
# Pictures
|
||||
if self._is_picture(shape):
|
||||
# Get image data
|
||||
image_stream = io.BytesIO(shape.image.blob)
|
||||
|
||||
# Try LLM description first if available
|
||||
llm_description = ""
|
||||
if llm_client and kwargs.get("llm_model"):
|
||||
try:
|
||||
from ._llm_caption import llm_caption
|
||||
|
||||
image_filename = shape.image.filename
|
||||
image_extension = None
|
||||
if image_filename:
|
||||
import os
|
||||
|
||||
image_extension = os.path.splitext(image_filename)[1]
|
||||
|
||||
image_stream_info = StreamInfo(
|
||||
mimetype=shape.image.content_type,
|
||||
extension=image_extension,
|
||||
filename=image_filename,
|
||||
)
|
||||
|
||||
llm_description = llm_caption(
|
||||
image_stream,
|
||||
image_stream_info,
|
||||
client=llm_client,
|
||||
model=kwargs.get("llm_model"),
|
||||
prompt=kwargs.get("llm_prompt"),
|
||||
)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# Try OCR if LLM failed or not available
|
||||
ocr_text = ""
|
||||
if not llm_description and ocr_service:
|
||||
try:
|
||||
image_stream.seek(0)
|
||||
ocr_result = ocr_service.extract_text(image_stream)
|
||||
if ocr_result.text.strip():
|
||||
ocr_text = ocr_result.text.strip()
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# Format extracted content using unified OCR block format
|
||||
content = (llm_description or ocr_text or "").strip()
|
||||
if content:
|
||||
md_content += f"\n*[Image OCR]\n{content}\n[End OCR]*\n"
|
||||
|
||||
# Tables
|
||||
if self._is_table(shape):
|
||||
md_content += self._convert_table_to_markdown(shape.table, **kwargs)
|
||||
|
||||
# Charts
|
||||
if shape.has_chart:
|
||||
md_content += self._convert_chart_to_markdown(shape.chart)
|
||||
|
||||
# Text areas
|
||||
elif shape.has_text_frame:
|
||||
if shape == title:
|
||||
md_content += "# " + shape.text.lstrip() + "\\n"
|
||||
else:
|
||||
md_content += shape.text + "\\n"
|
||||
|
||||
# Group Shapes
|
||||
if shape.shape_type == pptx.enum.shapes.MSO_SHAPE_TYPE.GROUP:
|
||||
sorted_shapes = sorted(
|
||||
shape.shapes,
|
||||
key=lambda x: (
|
||||
float("-inf") if not x.top else x.top,
|
||||
float("-inf") if not x.left else x.left,
|
||||
),
|
||||
)
|
||||
for subshape in sorted_shapes:
|
||||
get_shape_content(subshape, **kwargs)
|
||||
|
||||
sorted_shapes = sorted(
|
||||
slide.shapes,
|
||||
key=lambda x: (
|
||||
float("-inf") if not x.top else x.top,
|
||||
float("-inf") if not x.left else x.left,
|
||||
),
|
||||
)
|
||||
for shape in sorted_shapes:
|
||||
get_shape_content(shape, **kwargs)
|
||||
|
||||
md_content = md_content.strip()
|
||||
|
||||
if slide.has_notes_slide:
|
||||
md_content += "\\n\\n### Notes:\\n"
|
||||
notes_frame = slide.notes_slide.notes_text_frame
|
||||
if notes_frame is not None:
|
||||
md_content += notes_frame.text
|
||||
md_content = md_content.strip()
|
||||
|
||||
return DocumentConverterResult(markdown=md_content.strip())
|
||||
|
||||
def _is_picture(self, shape):
|
||||
if shape.shape_type == pptx.enum.shapes.MSO_SHAPE_TYPE.PICTURE:
|
||||
return True
|
||||
if shape.shape_type == pptx.enum.shapes.MSO_SHAPE_TYPE.PLACEHOLDER:
|
||||
if hasattr(shape, "image"):
|
||||
return True
|
||||
return False
|
||||
|
||||
def _is_table(self, shape):
|
||||
if shape.shape_type == pptx.enum.shapes.MSO_SHAPE_TYPE.TABLE:
|
||||
return True
|
||||
return False
|
||||
|
||||
def _convert_table_to_markdown(self, table, **kwargs):
|
||||
import html
|
||||
|
||||
html_table = "<html><body><table>"
|
||||
first_row = True
|
||||
for row in table.rows:
|
||||
html_table += "<tr>"
|
||||
for cell in row.cells:
|
||||
if first_row:
|
||||
html_table += "<th>" + html.escape(cell.text) + "</th>"
|
||||
else:
|
||||
html_table += "<td>" + html.escape(cell.text) + "</td>"
|
||||
html_table += "</tr>"
|
||||
first_row = False
|
||||
html_table += "</table></body></html>"
|
||||
|
||||
return (
|
||||
self._html_converter.convert_string(html_table, **kwargs).markdown.strip()
|
||||
+ "\\n"
|
||||
)
|
||||
|
||||
def _convert_chart_to_markdown(self, chart):
|
||||
try:
|
||||
md = "\\n\\n### Chart"
|
||||
if chart.has_title:
|
||||
md += f": {chart.chart_title.text_frame.text}"
|
||||
md += "\\n\\n"
|
||||
data = []
|
||||
category_names = [c.label for c in chart.plots[0].categories]
|
||||
series_names = [s.name for s in chart.series]
|
||||
data.append(["Category"] + series_names)
|
||||
|
||||
for idx, category in enumerate(category_names):
|
||||
row = [category]
|
||||
for series in chart.series:
|
||||
row.append(series.values[idx])
|
||||
data.append(row)
|
||||
|
||||
markdown_table = []
|
||||
for row in data:
|
||||
markdown_table.append("| " + " | ".join(map(str, row)) + " |")
|
||||
header = markdown_table[0]
|
||||
separator = "|" + "|".join(["---"] * len(data[0])) + "|"
|
||||
return md + "\\n".join([header, separator] + markdown_table[1:])
|
||||
except ValueError as e:
|
||||
if "unsupported plot type" in str(e):
|
||||
return "\\n\\n[unsupported chart]\\n\\n"
|
||||
except Exception:
|
||||
return "\\n\\n[unsupported chart]\\n\\n"
|
||||
@@ -0,0 +1,225 @@
|
||||
"""
|
||||
Enhanced XLSX Converter with OCR support for embedded images.
|
||||
Extracts images from Excel spreadsheets and performs OCR while maintaining cell context.
|
||||
"""
|
||||
|
||||
import io
|
||||
import sys
|
||||
from typing import Any, BinaryIO, Optional
|
||||
|
||||
from markitdown.converters import HtmlConverter
|
||||
from markitdown import DocumentConverter, DocumentConverterResult, StreamInfo
|
||||
from markitdown._exceptions import (
|
||||
MissingDependencyException,
|
||||
MISSING_DEPENDENCY_MESSAGE,
|
||||
)
|
||||
from ._ocr_service import LLMVisionOCRService
|
||||
|
||||
# Try loading dependencies
|
||||
_xlsx_dependency_exc_info = None
|
||||
try:
|
||||
import pandas as pd
|
||||
from openpyxl import load_workbook
|
||||
except ImportError:
|
||||
_xlsx_dependency_exc_info = sys.exc_info()
|
||||
|
||||
|
||||
class XlsxConverterWithOCR(DocumentConverter):
|
||||
"""
|
||||
Enhanced XLSX Converter with OCR support for embedded images.
|
||||
Extracts images with their cell positions and performs OCR.
|
||||
"""
|
||||
|
||||
def __init__(self, ocr_service: Optional[LLMVisionOCRService] = None):
|
||||
super().__init__()
|
||||
self._html_converter = HtmlConverter()
|
||||
self.ocr_service = ocr_service
|
||||
|
||||
def accepts(
|
||||
self,
|
||||
file_stream: BinaryIO,
|
||||
stream_info: StreamInfo,
|
||||
**kwargs: Any,
|
||||
) -> bool:
|
||||
mimetype = (stream_info.mimetype or "").lower()
|
||||
extension = (stream_info.extension or "").lower()
|
||||
|
||||
if extension == ".xlsx":
|
||||
return True
|
||||
|
||||
if mimetype.startswith(
|
||||
"application/vnd.openxmlformats-officedocument.spreadsheetml"
|
||||
):
|
||||
return True
|
||||
|
||||
return False
|
||||
|
||||
def convert(
|
||||
self,
|
||||
file_stream: BinaryIO,
|
||||
stream_info: StreamInfo,
|
||||
**kwargs: Any,
|
||||
) -> DocumentConverterResult:
|
||||
if _xlsx_dependency_exc_info is not None:
|
||||
raise MissingDependencyException(
|
||||
MISSING_DEPENDENCY_MESSAGE.format(
|
||||
converter=type(self).__name__,
|
||||
extension=".xlsx",
|
||||
feature="xlsx",
|
||||
)
|
||||
) from _xlsx_dependency_exc_info[1].with_traceback(
|
||||
_xlsx_dependency_exc_info[2]
|
||||
) # type: ignore[union-attr]
|
||||
|
||||
# Get OCR service if available (from kwargs or instance)
|
||||
ocr_service: Optional[LLMVisionOCRService] = (
|
||||
kwargs.get("ocr_service") or self.ocr_service
|
||||
)
|
||||
|
||||
if ocr_service:
|
||||
# Remove ocr_service from kwargs to avoid duplicate argument error
|
||||
kwargs_without_ocr = {k: v for k, v in kwargs.items() if k != "ocr_service"}
|
||||
return self._convert_with_ocr(
|
||||
file_stream, ocr_service, **kwargs_without_ocr
|
||||
)
|
||||
else:
|
||||
return self._convert_standard(file_stream, **kwargs)
|
||||
|
||||
def _convert_standard(
|
||||
self, file_stream: BinaryIO, **kwargs: Any
|
||||
) -> DocumentConverterResult:
|
||||
"""Standard conversion without OCR."""
|
||||
file_stream.seek(0)
|
||||
sheets = pd.read_excel(file_stream, sheet_name=None, engine="openpyxl")
|
||||
md_content = ""
|
||||
|
||||
for sheet_name in sheets:
|
||||
md_content += f"## {sheet_name}\n"
|
||||
html_content = sheets[sheet_name].to_html(index=False)
|
||||
md_content += (
|
||||
self._html_converter.convert_string(
|
||||
html_content, **kwargs
|
||||
).markdown.strip()
|
||||
+ "\n\n"
|
||||
)
|
||||
|
||||
return DocumentConverterResult(markdown=md_content.strip())
|
||||
|
||||
def _convert_with_ocr(
|
||||
self, file_stream: BinaryIO, ocr_service: LLMVisionOCRService, **kwargs: Any
|
||||
) -> DocumentConverterResult:
|
||||
"""Convert XLSX with image OCR."""
|
||||
file_stream.seek(0)
|
||||
wb = load_workbook(file_stream)
|
||||
|
||||
md_content = ""
|
||||
|
||||
for sheet_name in wb.sheetnames:
|
||||
sheet = wb[sheet_name]
|
||||
md_content += f"## {sheet_name}\n\n"
|
||||
|
||||
# Convert sheet data to markdown table
|
||||
file_stream.seek(0)
|
||||
try:
|
||||
df = pd.read_excel(
|
||||
file_stream, sheet_name=sheet_name, engine="openpyxl"
|
||||
)
|
||||
html_content = df.to_html(index=False)
|
||||
md_content += (
|
||||
self._html_converter.convert_string(
|
||||
html_content, **kwargs
|
||||
).markdown.strip()
|
||||
+ "\n\n"
|
||||
)
|
||||
except Exception:
|
||||
# If pandas fails, just skip the table
|
||||
pass
|
||||
|
||||
# Extract and OCR images in this sheet
|
||||
images_with_ocr = self._extract_and_ocr_sheet_images(sheet, ocr_service)
|
||||
|
||||
if images_with_ocr:
|
||||
md_content += "### Images in this sheet:\n\n"
|
||||
for img_info in images_with_ocr:
|
||||
ocr_text = img_info["ocr_text"]
|
||||
md_content += f"*[Image OCR]\n{ocr_text}\n[End OCR]*\n\n"
|
||||
|
||||
return DocumentConverterResult(markdown=md_content.strip())
|
||||
|
||||
def _extract_and_ocr_sheet_images(
|
||||
self, sheet: Any, ocr_service: LLMVisionOCRService
|
||||
) -> list[dict]:
|
||||
"""
|
||||
Extract and OCR images from an Excel sheet.
|
||||
|
||||
Args:
|
||||
sheet: openpyxl worksheet
|
||||
ocr_service: OCR service
|
||||
|
||||
Returns:
|
||||
List of dicts with 'cell_ref' and 'ocr_text'
|
||||
"""
|
||||
results = []
|
||||
|
||||
try:
|
||||
# Check if sheet has images
|
||||
if hasattr(sheet, "_images"):
|
||||
for img in sheet._images:
|
||||
try:
|
||||
# Get image data
|
||||
if hasattr(img, "_data"):
|
||||
image_data = img._data()
|
||||
elif hasattr(img, "image"):
|
||||
# Some versions store it differently
|
||||
image_data = img.image
|
||||
else:
|
||||
continue
|
||||
|
||||
# Create image stream
|
||||
image_stream = io.BytesIO(image_data)
|
||||
|
||||
# Get cell reference
|
||||
cell_ref = "unknown"
|
||||
if hasattr(img, "anchor"):
|
||||
anchor = img.anchor
|
||||
if hasattr(anchor, "_from"):
|
||||
from_cell = anchor._from
|
||||
if hasattr(from_cell, "col") and hasattr(
|
||||
from_cell, "row"
|
||||
):
|
||||
# Convert column number to letter
|
||||
col_letter = self._column_number_to_letter(
|
||||
from_cell.col
|
||||
)
|
||||
cell_ref = f"{col_letter}{from_cell.row + 1}"
|
||||
|
||||
# Perform OCR
|
||||
ocr_result = ocr_service.extract_text(image_stream)
|
||||
|
||||
if ocr_result.text.strip():
|
||||
results.append(
|
||||
{
|
||||
"cell_ref": cell_ref,
|
||||
"ocr_text": ocr_result.text.strip(),
|
||||
"backend": ocr_result.backend_used,
|
||||
}
|
||||
)
|
||||
|
||||
except Exception:
|
||||
continue
|
||||
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
return results
|
||||
|
||||
@staticmethod
|
||||
def _column_number_to_letter(n: int) -> str:
|
||||
"""Convert column number to Excel column letter (0-indexed)."""
|
||||
result = ""
|
||||
n = n + 1 # Make 1-indexed
|
||||
while n > 0:
|
||||
n -= 1
|
||||
result = chr(65 + (n % 26)) + result
|
||||
n //= 26
|
||||
return result
|
||||
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
@@ -0,0 +1,79 @@
|
||||
%PDF-1.3
|
||||
%“Œ‹ž ReportLab Generated PDF document http://www.reportlab.com
|
||||
1 0 obj
|
||||
<<
|
||||
/F1 2 0 R
|
||||
>>
|
||||
endobj
|
||||
2 0 obj
|
||||
<<
|
||||
/BaseFont /Helvetica /Encoding /WinAnsiEncoding /Name /F1 /Subtype /Type1 /Type /Font
|
||||
>>
|
||||
endobj
|
||||
3 0 obj
|
||||
<<
|
||||
/BitsPerComponent 8 /ColorSpace /DeviceRGB /Filter [ /ASCII85Decode /FlateDecode ] /Height 80 /Length 4282 /Subtype /Image
|
||||
/Type /XObject /Width 400
|
||||
>>
|
||||
stream
|
||||
Gb"/k$+*^]+31jd1_Sc48j,Pi+@:`R01h=9+]FPXQDmE0%*Lb4@[Wi36jU!;cssJbQ5,g%R?K'+$#.h<qu?Z`Dn#2Gqj`$$\bE9$XS)%of4Vd>cT_6mF8#7^^Y_6P]N!%L#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j4Z3bU/Gm9s<T86G'Ht,"?C)(`3j[rI\Y+2%=-fXJAVq>eJc&9=)%mQH&Yh<W#b:QHSf&hc5bR6(7?1RO+2U+,2j=JTJHV/n1*m;JAGbDu[IX.pg30l0S'.0*($<'u;[b/GRDJ[J=c-W0HaX>?jIi$uR^]%u?lJ6Z*VV,Z28T=.3[G"!N]2!6iqW[_CVOQZ9Um#Qd)&t%d!r@Y0=g5[M*c.,qcc"UaVkc?<W;kud8W>KHVDN2.L1\=-s4#kKB4PQPI/e#*[DZR^Y$^Xi6K`(0^so>.#p<K8pN:ein%PZ#Y7*R@MH861DpLo<$rl6(H_8?WBC8!u2*l:OF52PthFZhN<gX0$3^m^nt74tEo^W)lR[oP32CBgdV=rlCn/?MttP@YH\'hL=3DKjK>7hgY.jk*u=UQX(s.K*Fd*WL[eV0^25,*fS$V)Q-Z/Ii`SFMMDe;iM2HMUsg37cCLf8L+b)KY6rDWN#jR5YQ.1R1gba'M2,4kCR4578^=b\Bn#r]R"h8?'u=7,fh/#GD*$m:^@g'YaL,g&OD*(\9V3qM@J4qIp#mRhYee0oG3^KQSpS`k(a)%\\KrFo]NtW+D/curY.W3&31Xd58H(q0_cISK<s@\A@tq[3b*;)prprplo?NNWP"U[H"mT)":p2cCY.e)nGK>lIaB)]QNP9.mWE2l3Yi$/1lIFGq?b"J/%A'=e_4!7DM'qA.H+Eb6+$Wn7]h3.+R^:;FLJDKh6Z0V`KM>R1?7!q\hg`Vs6PqO%XsU&A'e9Y:\qjB8:9X.p&5omkN]TihU"VSM0Gdu%IekLW7Z.T+=gZ8?G+)N`8D1/:))EVV'V%>@o^?^e2`FI#RXkRcVk5<aX)Gb<Anp($Eo(tDRA['co9=J!r?\(k+3obVpgT(rh)[&!*=m"?fb<WboEW]&a*9n`H`'s&IkQkBo"rK"ncnu1$k!hAk*UR=.2DO6/^buFp=jJKa-\QsgioA9CY[S7mOaK!FH$moNAmYB\)0A"En%&3d-/qKIR!Etjt(g9!ZEncRkEWkI'W[)AEhgn9/$+_o'-r&1h\!<m^U0S31\a,_&Rd"r.'Np%gT9f8?\3>1O$"8"QQu<lL!5$/k#`#:KEulTq4-7/YE)\g=Sb$OCcUO)BHRhQu09oJWIZkkosFc_B+&7'c_j!")cW%]@9&7fl>'od45V^,W"[b:mfhGls_c]o[:8?WXO3sS%ABFnD/;VaJp_j5H4#BX4qPO#9Rd2UQp3!11,Hp.<#+W;VjHWUp(CD>tmalGRY\Uq<)U7[bb1;!ICRCfbRQSW`R7B!G^;uZXfo5`5U7D;,E%89G,#:1)%4QDA%S)!5IL>R;C=R4rV;kBDYiJMaSpYGjR@-K1X!l^X,hul.*@fk/SRgZtX.(?1#F1Z,3(>l3p>i<Pf+sbdtG`]h4ZQR\8ke79"3MknReA"?c^RDe@Fjk1.cu5MjEDpdNhJ+7mGf2KIY:S*Y\2Cs77pae4B]4nt\0_9hN"cX*)6Y95CMGu-b.h<l/f)o087<L+.ZYQN\"^nJd(2$BIfm6^Z>hKjO-5DXhZKVbK,R$;#1+h5rhZ?WW+cIfDMR5M:U;WJk^U1M2V=1pp;4^.2,/RU"b7N@$b8R4LOO?H"DR.Lf`L[*m,BTYmDZ_t`L-M$_)#8#p!)[O706GPi_l#Yq>cO^MHRc(Jp:hO`,H*Y]jp")!6$Iu21q$\8nLN&Ju<?TEli:_c^Fu;7mar?jW@Fi5=&+@XX3Du$Vp!Z:kp'-MBe4(Gq5273Z*<l$oQj,ndL:>:,=6H/*LPHo45Js7W8j$_!Qm0FH1P&^"`>@W4%?`Nma<X,sJlXF*,/9?-'cJp]Gl[CD(*jN88AiD,rcf:jl=)$?G1A+QH`L1Y,qGh381N!)?4VfakRqR\de*W_P5=i_rQ88,Nf"08ju'!L3:gtBn9tR`<1O'UehuL-ao(I9mcdD[iu:\EjK;,iTiXhVd0(hgkW_rte\s*ID1Wu(.MjQ`-_-:KRW+1tA<S?3r8>E^)_qfq#N;4tr+%k&Ep8k#92@_4?NnV=N@"8F,!hg:if"abZSI*B&dFMB&j8pk=5i_MJAeY/_a-bBH!b7VKr\Kt#C"Ke<_A>`"`=AC>VJ=jpNj/XAJ.8N&11/:hfIr$D^^R2#qRLKK:(9GU8"CB@_;$5Fq-q:K0TBPN]^2`GM'aEs1Y+T=D'>N2JXWoc8.%IYO^gsm'1RJSeGm+YDRQhLku5aKi&&h'k:Ae':8oK<la[fL\k0;fH3(LIfJts]t<l4*,ri:knWWe!M_E[M,&V9JH2`"=)ml_1[8!OOU7V,rHd]X#^@U_hK>1_Fu*NH]a>^r>**\J#14;Ei@8Dd[B!VZ.j64i(icM@UQ_>]1i+QL[q8@sXNl,qq<0pH2r<c]E5`R>K@bgt+3u4X[=5N,XXpe$Pa+h/i2Ns+!9@kBH_P,uQG__S.W7M^frRPr4EZHW;p0Je?#:'3`%IWs^jMgsS>TFs]-96.iKS'H_`---RRk+q]Jr]FS(In4Pq-F!6Cm%,U[%@0.OI2<<)q%YS\L]"SQrA8jisi-Yc]j.NcUR5eZO4@bV<6:Q<7Y8Tbc.:)0RB[f;uae0#hXi-F,V+Y7!Mj#7a2'<d>UX7up@?R.l5hdJ`J2qIRW9l3nLb6mCBmOi<W\odW.t='rA%`7sRbXB5/RD_LA/<@gLr;i'i3jlV::Z3F&:]ir"sAd&Y6P"h>gnWA-O?D51eitk>F^2j&Iq6CcN2Ju0jXH_V;7Z"7$/f.cVY>Mu"+'&]*\$$EFH_au5?=QCNV/dCcC.k5.]`boT#$n8q"$7k7cbB=_S?6!sI(ERNS%rY/q#(V&?"M=dPp_pD^a<mS>iJ84-qUUOnpsEBD(@=c8&j(fD<_iW8Y:1]3'Z*lk814$BMEn>20Z3q9`%2[odf^kVG8_KfBHJTq!iP-bZf!WUjfi-Z(mjNh$1Mk%I4bUXT_KbmDgHtQ7Z[%/U=`ol;d(+7MLe^9J.%pG(>?Z7,R7p2_!_Qbsrj-nZ^jp1P_<pXaK9!2C;b6ck"=Cj_ThjdJFfoo]T[$FN[^)H53%_>QETt1O4#d3il&h>-]FM?E7.3BltXHM-bbl_r^C;;uMGdf.Kh%L(0?a^%V$SMIKn-g/OBA,Ng_8qOt.G4*;07b-d&^'[LU$f5ngd%r-XNimO'c=1SVor0:Eg?<1-k=*lR5.^@!L"%EH/XBn&hq=*'_o;%t#(A>I6JN':Wh5=&pRCU'1C"15l6HQiH<#l)E>c9A33g31NEH\$h]o'o:W53E#msr(FBMb0g*jP1nCIbQ^<-?M19Kr3mq8.j:>;q*:p4Rb"@"DU#`i.DU&`=Vn-ANGOK'T46_'jF^$R0`j>ib(E*\_<8o*cItM:B3D-9Z>Of29HcT0]Z'G'co.PNW`2:qpYXp0-36TIRP-&3V+PPe>^kkuHt*7[/f`Z?74q^`DXV.TS7@]I@7J7#?[&(&hPL%\629`r50o^;oKq?P9#!l9@Fff9p3njK2nUHBg!&A`c[uXD61%4M,a"/_P#gZUo)#L[uI,Q>:BQkk3P?Scmo]DXk])TLK"NX2u"><@[CElgT\uF2.fcn<iiPL)@TrV2\AYDo>2%@`(OZ<M6L#'7K_ZStJZ)]&Fp39s]tR`?'J?rE-I11YEH*I?3FE.#D8]B:lU#l-Q&"X6RDb@GL2>K[lYeY=buQU?HWK8[#]q-;`G(]:<Ao2d9eEHWd<E81SK4JM$5dbM`T<KnN,YjlTi4>kV>d%&i?1&P=:i,4>V2MnI*kV+_s8='X"H,gcL;Uo:%-"-M]-mmX/gFJ;bSiNq;:Y3_r5g<a"7!Y]Bk,;T:p3c2CBn/b6lYENkm?LZ[fW1tg10cT`#9kR&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j46GWU$GHRA>~>endstream
|
||||
endobj
|
||||
4 0 obj
|
||||
<<
|
||||
/Contents 8 0 R /MediaBox [ 0 0 612 792 ] /Parent 7 0 R /Resources <<
|
||||
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ] /XObject <<
|
||||
/FormXob.0315aed9f6006a101b3226a3b7404028 3 0 R
|
||||
>>
|
||||
>> /Rotate 0 /Trans <<
|
||||
|
||||
>>
|
||||
/Type /Page
|
||||
>>
|
||||
endobj
|
||||
5 0 obj
|
||||
<<
|
||||
/PageMode /UseNone /Pages 7 0 R /Type /Catalog
|
||||
>>
|
||||
endobj
|
||||
6 0 obj
|
||||
<<
|
||||
/Author (anonymous) /CreationDate (D:20260126172022+01'00') /Creator (ReportLab PDF Library - www.reportlab.com) /Keywords () /ModDate (D:20260126172022+01'00') /Producer (ReportLab PDF Library - www.reportlab.com)
|
||||
/Subject (unspecified) /Title (untitled) /Trapped /False
|
||||
>>
|
||||
endobj
|
||||
7 0 obj
|
||||
<<
|
||||
/Count 1 /Kids [ 4 0 R ] /Type /Pages
|
||||
>>
|
||||
endobj
|
||||
8 0 obj
|
||||
<<
|
||||
/Filter [ /ASCII85Decode /FlateDecode ] /Length 260
|
||||
>>
|
||||
stream
|
||||
Gas3/9kseb&-h'>I`6Z84fgHmCc;"L7g6_e&889#h,kA$Zt,m0Hdcho6>O[sLZ+YF+:QDRLY`5CAhdUI=MeslW_fp84Bms2r(UspMdQW.jtWA9rW?q[M1*5b[XIYc1kOQ$55sEf7La^q2$a/'T.)S#<V#*e,['$SVK^(f9:,Nq;AW\a?Zt7p:RM+pHF)-4F;E;l5ui'$5;T>HA_.,@?H2a/)Ol=NY+4r->>:n6'/ubPg6GC78<Gb)GJls9>QKuE<U0~>endstream
|
||||
endobj
|
||||
xref
|
||||
0 9
|
||||
0000000000 65535 f
|
||||
0000000073 00000 n
|
||||
0000000104 00000 n
|
||||
0000000211 00000 n
|
||||
0000004683 00000 n
|
||||
0000004939 00000 n
|
||||
0000005007 00000 n
|
||||
0000005303 00000 n
|
||||
0000005362 00000 n
|
||||
trailer
|
||||
<<
|
||||
/ID
|
||||
[<5d5eceaa0d906ef66e559ebcd616f18d><5d5eceaa0d906ef66e559ebcd616f18d>]
|
||||
% ReportLab generated PDF document -- digest (http://www.reportlab.com)
|
||||
|
||||
/Info 6 0 R
|
||||
/Root 5 0 R
|
||||
/Size 9
|
||||
>>
|
||||
startxref
|
||||
5712
|
||||
%%EOF
|
||||
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
@@ -0,0 +1,79 @@
|
||||
%PDF-1.3
|
||||
%“Œ‹ž ReportLab Generated PDF document http://www.reportlab.com
|
||||
1 0 obj
|
||||
<<
|
||||
/F1 2 0 R
|
||||
>>
|
||||
endobj
|
||||
2 0 obj
|
||||
<<
|
||||
/BaseFont /Helvetica /Encoding /WinAnsiEncoding /Name /F1 /Subtype /Type1 /Type /Font
|
||||
>>
|
||||
endobj
|
||||
3 0 obj
|
||||
<<
|
||||
/BitsPerComponent 8 /ColorSpace /DeviceRGB /Filter [ /ASCII85Decode /FlateDecode ] /Height 100 /Length 4720 /Subtype /Image
|
||||
/Type /XObject /Width 500
|
||||
>>
|
||||
stream
|
||||
Gb"0VH#OJ:qoA4A'Hn[Z$4u82K`j4ZR%PRX#F(qaK&V?K8<b&(;#dar'M8Oni!CfV<*>q$(4m``.P7J2`EVk#7#VtC6:(*s$)76DC7esMC3FfEP21e=J%f9>Up;d>4l&7WcZJp*DIk*ozzzzzzzzzzzzzzzzzzzzzzz!!#9dqY/lsQRl8pCtPt8mFiS/o[0Y;WL90Bj2R)5Y[N/Gfn0f!(^fES:Hsio97D?(Xn?5jq!"]KU?6X;&P"ZofW]f$p(q"VdAg3IP#\)UWg`=MO$9TD0>@3js(id(lnKeb'nVdJ7e;beRl:iu3cs8nIEn"+F\0s:]mE81*hAmoIah4b[;OfHn`%MZXAic<H:=+SS^nBEXOMIOI2D3Y<1k'jGjr"Mb7m#8djq#sB[J"W0CSVhD_EOgna,8=^]'+;Q/-Q29o7K`^M3`IrI+P7M:JgD-;8@g?QuaS(L!S'NU#&pVmQm&;+ooep7=EoT&nD(?bbsoCn2i[<[UJujmhj_M<u#m'ah(clITBmFV`P'C`bB@K_BHIO[n\%kN(^*?b\d]&LhJ.U*@&CM]/WZY.jbti9?:_k*Y>(J)7jH@XF(Q5('jYY[u"DM\[nu[VacM!sePdg%40X+-%@'f%P<0baG`E4oP$%Q/JgWmY\Um466eV$?Y)Q9)nh\c]Mq3f`',ShaFWth-oBcOd=q3cTY"+4B:C5D=,9ZUiS/hg5kVA4*K+[1fUQ5bbN_se_nC2++F"$]j*.j.s0:>;7?2OB:nigYB[?_a,Ulb<cmfe?*!8BYQ*mgYI_'G:#:9hc!%'nIu9\#K.OFt>Aq1i\Q1k(X_QRsW=DbSc%F'bgtZ/2>e+n:Kbn'oJ*sr;^;r/1Z+Vo?T!W4\7L>qdS!IH-WeB#r>ch5>%TRL&'caJA=^o?na4Y*tXgMf5H"N01jlPU;HhZ+Fp?gW#Ip5H[Y;u'ao8X`t6%]C0Gr*LD?+V"4C6Y<]^1GKRW1+$NV&M@2Zk6GD=d]J*qS.+7cN!h6O!e)cfIf(11iD*Y7*8G.!Fu"f5Q5o`Fk=$<gK"E?j,ZG(Q<S6H:\dHiQ8`a=>Zb+\X]m_AEjKB&FH%hV\FAFmKDpQr8j8Fc:%F6j4nD>GFYUI\FW`_flD4$L7>hqmCTh$Uf"^_*P&AuHA>O*GCsJP2O$EX=EQ9*O\8c#4&JZa4ARj7@85!.BB?cmQFmIWVr;9Tt>%neNS8ud$:Htthg-nk9OnPg[;$TI`>0jlL=-.a+_hJV!J^kl(gQd$+PUS\<me#jih7@a_WGZZ(.4K"nV+[.iR&jScFk0\%2K,Zhh0o%R)Nq/WF?l+\*Dd3krNhA#g[-\oC]W"'hnHd4_hLeQjHEBn&n644.4m-ZIetY!]Maa6"3/b.DR^j3BPL<$4Z,)s+'5XPm7A'P[OZn]2^N_0O[lEQSs)o18I6(3r!Qc+D!gf8aN/&O]Qq2:ob7jW;;79<m7q;t-1bA`T7?icP9s!jiRtHa:-1$`1XkWli8@t0Uu_-o6OtUq>"[U''if?a#-Wq5c7%ag\H8!"ALF'oU?-RQD7B?-GJ]">g6^'>CFf@5s8D[^<m$#K27ioe<`YMIH[l(oGML?\W`P:J[(9Uijd"PR!k=->Y$F-4CB"/,6\Z#s5Dlg2HM"F7VJkA+mbDo1>N-;k32'EW?6)(KY`B.a^\mY\PAK*gH+(7uU^F-po`(RMK06EPHHdD0;Bn\le,c[Y^V3lFa4(IC]mFtt!K@dP<o88m]iqJ+@)2E/k&hb!@XF*Gief89[^t6$$47L+p[?u]I+odK?/mT/=%`2+)fO@AogV9q*r!U1m1g?N]&Ti<e"o\RXmOcGj);^2<k\(R>\obnltl>ED4q/%Sh>o`U)Q4>YWf)aUl].\e=Ep??@2&sT>Dj(+.kQ]\905L.Bs_jQMg@#5Ad*SY8q<&o<LoLF#$U&]1eeYfg_3L=pCsBn2Zo8@8IZ+Xg>-8CKR8V]"=q/;I!IC.2@XQ-aimNpYWG+0>@4U4t=ghn%S,KY6Ikm?-DX7<>iL_CI@3\nVAbo+bpOJCAW-_HQp]RX&=4gH(-^/Z@s5UCp8LSn\c))=o$*Q*<M^D^#4JMJtt=95Q%_uUo1-F7q-h)d`H"f/Jt%*3B9)\WKo2Erql0!qemE![bE'dH=+t)O0/c#?9IC9XEPfCe/A/_qsP1I:Gi93mAc2E?t738#p#IUW/S7Mm!5=<`aIfEM<Ys2>e&.Y0ZhJX-aq'tbOsIoYE.k=J%d;VB:aB<bI_rblEfBj^p1R=K*Hi'nV;J/+I,YT[]<3iSq0j"_fD5.GHQ9sRi?Hm3c3S-p!79pR,Q`;q!mC0XK\qU5$G/?oAYN4XKK9![O9M9;(JK$g@P<;orgiE)Wd/_e6&iR=?W_Xldsm><h,SP+L,3L,ntesb+_2WpLI=+=Q*1F"S(>qmjXiPmbHLOYe%@qT_T#OK#H+:rVJ+)kHmJLjHDqC"3ei1+I0_7b&fC4SN?G-:Hn<P`F4_m)X;X7>C^ac96r3OHOcmBeK9_BW&gqhjl7$/j4:&TqtBm]_@&#AP,Y']*Eo)'ONPAD?/7inL-[;Y?u405b]Ka3.k@s]eE(j,^\#rI[BQU.aF>#4B$F4/opmSM*F5.or8NVf4NVD[_hmc;1iLl9739e\*dBrn2.Z2*I#rOpSJRG>K>mOaX&`@A]5&)7CQehdOsNbC>B_D5cTCSXB*di92jW_WX!=EhTKNR%fVt.%Q<$j[iSM!0d6/k`CY(3)+XeI\r:.i,Kg1O$IGhnlT&gm[L0VDFcUFbqACJhEm'4S\ChfI]q:mQ"ZL[OBmJ_6*\'6h<$$-W"[^F9VSm/"@Ys!-Y/kBOeN9p]O$ui,lR;]Y'fWi?-gh&>)baIM5;iRQYFQq5Me#,uCc8P?h`0K;LQ;G)X')<<ZlIDq&LLPU^bo=&gcErEUfq:W`I+ft!GF'4,DQL*1IX_:-FmGd!pPJ9q(G?8P`t=nl4**0bl/9C1$Pk;?WMl,4lD^\UVMQ6bn$qBfs80*^[?ET$I!^-aH!XgKQFCSW7VR7m>c6G0oT/C6@E@rs_sQ]md5bQ1:uFPQQE5J6oF@\//u>D@rl6i0Db`9"CffLREne*h9ea#&lL)T6cQ+8d[]`uKp7-3LZ,`(u?(+fr>(mI*G/\k-YJ:uXP.0=t4*2mZ-eQ't.M\@&WZ\RXWp+0AS>cW33cqTe`:fXp?:r\D9j`52V-"$nNukEh^\mZGUTX8#HKq+^Sc;g;39(Dp^!D$\>%6A^eKD`,3r`Ehh.Y<QB$Hd6Dn\4V,K$O2eQ#]H'IHuY<'.PWh7M8sFIp`W<?e^(pf-sk`:cWX(0Rr5S=GEL-Ru1K?[mL]^3qom@^04`'ab3DaOa@KMi0rX@XE^O>K:6c7M&12fk$N'7q-hiZ)75?UbZH"N)8kd3WGbMX"P/.K^RR%.rqaT@h#u_AEs1)jILMOBks:._,[m*eQ/HMh/2T8\^k5p-Gbk4:UO]EUnspPj)`O0k>MR,M8scu:M"<+[OYCfCtV].VbWfJfu8q0hWVoO!s]=g_kY<KG3'BXc*o(K]QH6CDr&"TSejAI&r>p4`r`?FMo`BP7H7X"Km<=Xfhj^&%sjRIEf&?W*]uFIg@FfT]->7T*G\=GA%PJ<HdnOVSmLf.+KHmSZ$jZQ*BecC<(3<9grri,I2*)b1:SDnGU+d]Xg6MY8$T(:.4?Uk`u[BiGe0Oe2f<H[Ue1=Kh4riO*S$)8!@o+s?;W"![]<f%`Y5^Zc;'ok\LTFOfW\3I*DUf6[:**:Q92N&d_']\[d1Hqldmd(IV"Q.V:G^=4P&F/.]W6G(>cB1O#e,SV5<H8f\>>$gU<+7PBoDYDti\Up=RQ$_FGu!./Yu`]tGE[]1Xflr3@X<>TH,QPDG[k\(f5EN#4:dBj9Dtm'hE@kFId$O4XJRs#<fi\gX(P(\HEsYBB9>>%h8.(c#WXs*h$)D\#t'aEg:?XOs\SA`#9^5CU9:p<JsU>L#D+>hdhTPki]s+6c*j=3>f=V>+D"=D5gHfUbY*f!X/5kZq(aU0s.TSSbiDpX_$Rm57L'='`/+(t;Rbo#i[rD<hl-MMd9XiU7<R]U8H\\)5nGm!GTqIWJoT^k&3K2m'gnqWfVra4/mgQ`:tY5PX.=H\ipm,paod-T=!9ie-6^rsOM%b,=gWO8LhMekGg^s513Ue4#ZT>@pYrm#1Im](Qt<AVg4pp1hYAJ<c+q=&_b]DnkJ,HRr<'>$A+9]t/CS)@C[llo,Jtei=]'$rSMO^!toPHeWb/:-J8LrU&LWJ)\^W(Lh_BC(Sq=]a:sW(9CiU>+L>jbY2:SMO\P;Zl(Q*_#4$"rP+Wa'D/kYlP9isqjdB"Z4^4CMs\(rfP=ldGLb]=a4-[40"Pb'F3QT9K-?3m2,`JlHL%\1Ij*8m=o$^)IJ``GG>8o)=:i+tB%*VOA&aJ4rTY`4<mp,AAS#lUS&fu(ONL&D/#q[E-aS3rEpZinTItFX7`LZA;mpPt<aK*M4`L.JdGj.pL%3[B<0^9=Js@ifcC6aG'JXr;^#lG4Z!G16L%!Kgc\r_t,%&S@[KH;Sbh`b-,X:&f!<?*P^juT/EcNB$W&AtP0_oZ%$360Lace*-_EY$)IJ\1l;I3ZnIJS%;Cu2i#mbPJcDt*f-eZa,X:4"Ls6%]AE=]p#qGtjbd%>DQu\n93U_d,;'5W.rd^OPtDft"Z(lDUVVUibkLA^[AGcHdpAzzzzzzzzzzzzzzzzzz!.Z!\J#>u+ci~>endstream
|
||||
endobj
|
||||
4 0 obj
|
||||
<<
|
||||
/Contents 8 0 R /MediaBox [ 0 0 612 792 ] /Parent 7 0 R /Resources <<
|
||||
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ] /XObject <<
|
||||
/FormXob.41b05a9cf8679f0fe6e7c30c9462b767 3 0 R
|
||||
>>
|
||||
>> /Rotate 0 /Trans <<
|
||||
|
||||
>>
|
||||
/Type /Page
|
||||
>>
|
||||
endobj
|
||||
5 0 obj
|
||||
<<
|
||||
/PageMode /UseNone /Pages 7 0 R /Type /Catalog
|
||||
>>
|
||||
endobj
|
||||
6 0 obj
|
||||
<<
|
||||
/Author (anonymous) /CreationDate (D:20260126172022+01'00') /Creator (ReportLab PDF Library - www.reportlab.com) /Keywords () /ModDate (D:20260126172022+01'00') /Producer (ReportLab PDF Library - www.reportlab.com)
|
||||
/Subject (unspecified) /Title (untitled) /Trapped /False
|
||||
>>
|
||||
endobj
|
||||
7 0 obj
|
||||
<<
|
||||
/Count 1 /Kids [ 4 0 R ] /Type /Pages
|
||||
>>
|
||||
endobj
|
||||
8 0 obj
|
||||
<<
|
||||
/Filter [ /ASCII85Decode /FlateDecode ] /Length 250
|
||||
>>
|
||||
stream
|
||||
Gas2BZ&Z[T&4Ckp`KUTrY_02PMb#<CFN=Wfj',kM@19sp55uUe"pptDD)Los"F*-#r%7t"K39EA8f/'^$OO.*D:jQe'n<f:3Cq8'p9Rm8qll,u+[sQj[W6hrFQL%\7G?"sX/%4LXYeUkIBuT`A)Y3?=ouE3GIShId3E("2qqVte.E2,r_bJ%q1G(F,@9C<XiC-L`O1W5it(MP9X]^nj..r=,_#ecrj!ceT&ATWd4)p.7/d!C@/gP%;p#~>endstream
|
||||
endobj
|
||||
xref
|
||||
0 9
|
||||
0000000000 65535 f
|
||||
0000000073 00000 n
|
||||
0000000104 00000 n
|
||||
0000000211 00000 n
|
||||
0000005122 00000 n
|
||||
0000005378 00000 n
|
||||
0000005446 00000 n
|
||||
0000005742 00000 n
|
||||
0000005801 00000 n
|
||||
trailer
|
||||
<<
|
||||
/ID
|
||||
[<38bd217c814ddf937f148e537dce51f8><38bd217c814ddf937f148e537dce51f8>]
|
||||
% ReportLab generated PDF document -- digest (http://www.reportlab.com)
|
||||
|
||||
/Info 6 0 R
|
||||
/Root 5 0 R
|
||||
/Size 9
|
||||
>>
|
||||
startxref
|
||||
6141
|
||||
%%EOF
|
||||
@@ -0,0 +1,139 @@
|
||||
%PDF-1.3
|
||||
%“Œ‹ž ReportLab Generated PDF document http://www.reportlab.com
|
||||
1 0 obj
|
||||
<<
|
||||
/F1 2 0 R
|
||||
>>
|
||||
endobj
|
||||
2 0 obj
|
||||
<<
|
||||
/BaseFont /Helvetica /Encoding /WinAnsiEncoding /Name /F1 /Subtype /Type1 /Type /Font
|
||||
>>
|
||||
endobj
|
||||
3 0 obj
|
||||
<<
|
||||
/BitsPerComponent 8 /ColorSpace /DeviceRGB /Filter [ /ASCII85Decode /FlateDecode ] /Height 80 /Length 3374 /Subtype /Image
|
||||
/Type /XObject /Width 400
|
||||
>>
|
||||
stream
|
||||
Gb"/kH&U9Q'ZR>43%?O'&jT85=3qe+N1f0n5S02hEQ*VDL]ZS)-n8)hB\WWtW#6!n.\\/*',>!9/rH$h#V%%<D1W$*'jLaXkBH(D['+neI4_Y@WVLF\J+T;ghR7Y(mQ%b#VGJrM63n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&/S]&BhHiT>FDK@e$Z6UXYgOsk+@,I0-;pc?Ln*mjmTTlQ+?K]e$#D.eB)O5NG7.u<*,RJ_p,8ck3p(&R=G"JnR/3XR:fF1m&!LSDdTjBc:RULFd1\@RVR9Y?IMP'Ajfu#dnfOdRMIRU5?(oCkRY2lY[PsI[bY!Lo7-qefWg9>U_8tF1IHqd?[hMs%mDBpEmdN`p:`8WC.(r[:;%"2F3\c4Mk792G,A7iab,>+^X:Q1B)Ct4JU?`lHM9=Q*\(H$'0?&(T*=du53O+iUIN?O=m=In-UqEE?ghWKl;bQuY*`RG[FO81(L.M1]2c_&%<`*Rq.JU[hs+4A7O4D[l.+Bm#OHs>Bg2RPN#ZQX55r(P>4!naFjcC+c`URpH'c],PFQUH`f2c]IHB5B)pF^[%Q[/+G3L2gM0I)]FA`$/.TW]/!sdN]/(^i(\!C)"CY$!K4SDm&fpLsK;:U@[oi*BN6O,DaRecA5ZZ2bqqSNigQne<3\(m/F9af2d;H'i,rO4%j7$89^HKB/%EH:d:UE-7IC?3kP08QNWFSkIn&tXU0hVtf\g_heWn%H2@\EYR$R+7l,it!qkZIsA%.8%:<b4Q+4o7Re<@/=uc<D?1CkG#t*Sj+#k=3?C8<eoR]gMmuDKs]#U,%B6^\'Q>l.UaQb$n85YGMOQX1#>:k7>o*u_[[jr:Hg5Je^Y:RKsDj_5eqt.H?@qPba^,b_QC<DQE1;H6P%jjc90Rgp0)'S2.M@\$lIp4C<@5NV3A+Sq=@L,V9GDV`O9L@%f5,%,!P;IfdpJOYbPun*G`61QOD3'0Y6GmF^2XsRCVfROl*/ge#f)W1s!?*VEG*0."1`MfU_Qj-_HcIc]o.$p?:mMIQ<MK^AHukRr/l=E7?<+:@V_=m:@ob4Q**VPHYWj[Zo=CRr=V!*BT#C(LJ_:?.qW/l#KCp@t8d\[H2o7Bt=PmC=",eKi0L-.*!_c/%pOEK43<A[3H$o[#=noHdqYr^oBKc5fjO>LsH:W;>mQ8`hXnJRbG389AnU"B(jou524Uh#I;&<U<No:XESS[)bnZpSA"f;hd8gI+P]H$/\(H<O[nan=8V^RLa-H@:Xf+/INI%?Zd4q"#jj>q*.4u=YF[nrO\>7O#of";9>"U0op9fI&cNL!f<:OL7XG#UCV^-W3n&X!D@elVYI"hK-(KVsoa8aWL<4F`IGb_t#tRF<8C]_m^L^GO6G3G5S09nd$$CBb>0uZ(bhmeXG1t'(r;5s6L_nY]J6S``aIm.70L=toNMorm#gcR5B07$ZWs&l!IuphJhN(6@caN->pY971W`M`H*G1Tioii3W&eZ2TZB^Na&P9Djoa9RTg9i]Nf@JZ^+U;<o?jn>70)4?,Z*\6\f"L%[`BJ6Kj9)%X!(RqAJ!s1(OmA61d_0J*HI^@ba8Pi</l@aVqWUPab%]CO5^)(hH3sGpX-^oTZd5(_lb]&C\k)B3IsfnOdI_nbfq2>Pl-9KMgjahN*A`OY%3@&b90[&bRjRh=*NU*X?K'!LVo@=gF%20@?O=gnO^q+Jr=$XI<l#nIRPZI$e8J>159G:[l)_5.-%,a+qd/.CDY\U4-ibD!])@[SqAD!$G2,qi7HmfIX"M?gq4^a:eU]L7T>7]\Ni+_8Hg7U"jUaRtKJ]bt'P<e&53@l:okMAKlUSY_?MI_!>3`l9op5MT]g<,Eau4CBfS9qgs8'hVO^q,7ISl%l9Re9VphQHBQpHf7;ouN+#41(!]Bpq.a"s<1RM\\qcK'V\`Q!N1@Y46f"8:opL%80Oi]-T\^Ju'R*bNtSnP$N;[5T*[5i*NZi1b"eWK@pn&8g.BWL$qSS5iR0*5>0K<j*&mC2tplBiAcJVg9(]ip]PZ1oSVKVeSV_/S3Pg9=ab"pNZ5V'2SCk-Vb@KNuhlT6mp3R?8XX`:Q.L9H,UNhF3YacG3W(VX0*uZ8>4e?>K_G;Rn$CUR?oeU93At-Arf=,=bsA0p$p(CN!F<.%bX@`pKfj\]c&XOS1!:`do;;tZ6cVZD;+'sUf$Bt,Q7PD^1rYmq+$QG]u!IsY+87o.os_h#/VHUYNfDXl;b!f:?gYC^>D*J>\&T\cBJ-sI_$N&=^rPk>LSVt;>P.46N:frEgm*GGj4=<qT,Y(1Zc+Z]h7%8,[8^^eLRgoB$=^7=$"Xl3[>=b41/JMmG;,"j2T%QRo@!%UIU4G6f=MZjM:Z2<iT63Xu]!qai:DQ:RWOH]Hn]!\#0inkU)(>L9M]+g80^DM^g"!X,S'0pD;8jH/WctXls@'Zr*MP]h7%8,[8^^e]B13k1X#5g%F[m.h'jn0tqYq>l-H\%mM%:Cju%L]qAkq0pc,X(bDM0Q0YGKF?F$>3]@tZ>A#n)hKGBreCD[e6OjG&;>-_QZ2iafV]GHA*H^f/E.J;%Om]_HNhJJ%MZ3l1UbMX*]CbunmMSLel=3ef=.@Tn,XYJoeZ)Vm1L<aCJD&:+Ijur)RA66H:%K_'"UbD-=0FNC4t9b>F/i:Y!Za@[WkasO461hJ%d@!MX3Q&p.'mXeE"=VfNAkB83#nEb-?5c%',NbTmp@Ic`7),9GbJp)qSmsX%1!Y73W6=/j,Gsc`f7.5:+Yeeq`Ai+!fL?NOM]56n(<,D$J18;P:'SXB'[uZ2^6A$<-j66f6Jk.&&c3W_A9(cRi[QK19NKCB\1b$_[lLN+.MEm1P"`E)t++5a[#+RZP^-(?@hY,mDDVfkSlm8e\\@>q0l2NF1QP^9\%[*ac\o<7!Y^`e%EF9dL)U==0$Z?$N0G;m+M)j+f_J:*uda;lh'4kU0Eu=[1gfDqKnase*WU=mVh!(:]"OUWP//]CqZjE&P4=Fd]4EP,j2"jQH?"R,$k%R$h6Q3^%g%3\k2'Q2t#0eqVHl3N]f0YdOSV61-pC:UfT.\lB:IuJp7hZ'6FlP!LEm!e%`Z.s*j]KJu),bq<-MI+2hQBC-aRH=2j-B+b#)kL'"'pn0^RRn5.!9II/5DmNHR"$lgsSKY-\*\*MgP]XVFu.o`0C'fI6ZKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&<`?/!KNiek5~>endstream
|
||||
endobj
|
||||
4 0 obj
|
||||
<<
|
||||
/Contents 12 0 R /MediaBox [ 0 0 612 792 ] /Parent 11 0 R /Resources <<
|
||||
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ] /XObject <<
|
||||
/FormXob.b6d21e33426b982eedc18c3d4e93428b 3 0 R
|
||||
>>
|
||||
>> /Rotate 0 /Trans <<
|
||||
|
||||
>>
|
||||
/Type /Page
|
||||
>>
|
||||
endobj
|
||||
5 0 obj
|
||||
<<
|
||||
/BitsPerComponent 8 /ColorSpace /DeviceRGB /Filter [ /ASCII85Decode /FlateDecode ] /Height 80 /Length 3746 /Subtype /Image
|
||||
/Type /XObject /Width 400
|
||||
>>
|
||||
stream
|
||||
Gb"/kGE<NX(WT`Jj\uYAo)R):e01]`";Q!n&d*n+.):?M$;N-L92SQib[Ni],#Y)4BR+![6n$6QZkR#<71h=R.UYRO+N",lDr8pJ4*g7;J)9%2]0:`2oQ9iO]^,RBHegiOad<J[KFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgJmN6a^Irj:'BVG/#9rV!+sehf4Np$4uGT6?Z%jn76cYI/bg\b0"PYFheo17N)h[bT;Qm:q@c230t>rr(GQpkKpmDTn_bcA]O+Jd#cU@+2Zm>]f;6bs;T&C"&dtb&Y8uEeu<LE-(N92GKeb>4LdJj]\*:qP]'A(VpnpR/2-8pYO1I<E>P5N\HYDRFS?rdSVTS(Rm^Cb\t8pp[.ji7rj(S`LM@bl-r9u\(u5id79+1Zkc$%??stuVZtblhmB,pZt^mu0`%Ls1i]8CHul4%*Hj.6mVDOM9@Q>X0"[KH505<^W'N@NK#fUn2VXT_I7/`G*I"#V]/HO(ZsFbo9PDDePMK]!H;tTT$fdk/bb^Xe<o%RJ;c=p@KgDW9>;u04LW/P]CXtHuJmX$+n(Vc<?4@i#F^)<je+N!;?@B4`(2I>\^&%Tk]_oPE2P5G57Z;<354V60?+`jnL(BV81&C9p'qn^>huY?a*aVo\^AQF(LMnkjY1\5I3=GilDM^Mf#@3^HP(^=%G)"Pl2nSAIgiK?0>KOPWqO$PVHF<:_o#LfLahWdh*9(3^m/EKhl,$Q3cLgQYNON\9m^^C9op:#?rd<2$Vk!&)d<th.i@gbDG"YOud]6@72asongem@R4YF`QZs^cCb&Z^>EY^Bos&AIDEpB'*`7%OW#]/JaVk$ICr,?V+Pq1smN.e[Kbq-r/47c-[@E9!u4pDp'F`gCN0YN(T*.>3lQo[*tf$^DS35K&9pYYmC(C!8KWI=YO-Yh<iYq"0n-PcX/R22T"lAiUQ?;[;g"V[q<\)6WG:J^rp+%ZAH>K@cWZ,bpLf<4]8ob:WF?31lf84UU8ba9QV_YEY=:-f*?M'pH:5PUm1J)N_lEV.S5l4J>"ICf>9o#Q>b\(rC/e:S+@s,o'A+RfKf[#p[C^,r^GPUR4gNZ=J]6kHiV:DYI45Dd,\<-li[]NSt[l+63)OsT7L1W3eYFAoO;cK:jZ>hAN$F,_R#nIAbhgQ+SUFQjta@9[u&(LAMenD)m[`R55a*Y3l"))/k=oMU)6$+i]jVa*618S=TZ_GFf=XBs_%K:K'Fo]DciT&f4e(<2n?Vtb+e^g%Nu,1VYfeOc,e:Kg5KNLk9Rcmq(6W8>+.5UZlp$,%&K@JAXf9t/kp;B>ou]%HuU9<jA3FHc!7?g+9s6HK-!P:8IcUIVq-FJPOLMNgE:%ADX":FE$rF][b6ElT3'f8;d%Jo:3eXU\iUJe7VBlWsY`p!Z_)<M"V>Q7>RO1)%6mY+XGj^Cfi\llJ`iigb(cAMjb)$@^lQ9+"%O3RN/\G-)Fg+T0@+7sBYYOlju6E^l%O_dec#[W(oiP)i[<g)DQB?>1+.GAD<*#ee)n]ZlQc:X6"mf=0EeR3-\Rc-pb@okO8@.<a?P.;n:M[q+PD?$2E+e8+3J=j@AkSTd+T3ms/eoJ)7>\^lM5KERoAErtC0R56,oZN&WoDC3T.FfcQ):>]e:aVd1keQ_tITI3"f#+HG?-@C\E:ct,l:h<Cp?GYt&rHEP%f@DuqiK3-OdO5;eandB'^*u(E>'Y7/kYTAclDW&K5RRRs"3=c::U8&qE3Bk7N4522.XFX-a_8A&BTV-MqW1^SOa5rC:q\>me%lnUEUg#`JHMa+Iu3uP#:.&O#=ggsUgq2ho1`OG(%W"^SARTTWNR+&lC)M$Q`-sKP).<Rn<-H*Y^_.@YouJ.NulS]BgMA@p*ha_HBliRAPSEe%/q,5$rS?H&(.ebb:^u]K\M$!o#]`(^ABPX>'>!(C!`PX,)AUtl)+6md<^LC%`&<0kH\Z:!VH;uD<4`a?Bq\X!.ko\0k6B4?aIal%rsfT"YPIS4"n?"LH<iq4aDoZS1+2c#!%I<oORlE.6243F,5XrA7$Vp<=5I%YtpMPkuDakPrW:M755EjC1M+ASW1"l%6ssh".hAcrM+K$.)r`aK*P&Hs.u$/ckS5US2A?-\M(ZV;>Fn=!Y?$@qsJM8hgJQ9F[&m!?Bq\X'Z@O/_1jZPJh0WIlPaMENKSCZH_qf=)`E/ta?<^&&/;eKN[POQ>c>>*@<Qcq<skZL!t1i)YtpMPkuDakPrZ,@m=)4N13gIaPe1(Bgc3F?fe]L"FH<.+3eU/;OU:9l)jAj1eZ7B0jUb:P*^U'nk0B7LJU1=RVTVD0$LIB(o)AMa*Y*&F=l%,@f3RrO7lp:h;as9gS_OV%PBlb1<mC&X1YMQ7Z;Q.M?N'DLPOF@X4;:2e@\4k)e#VPa.Wa&'e[fnk(@eW9s8Hp3(LD&9UR,/A3p9VHP#n-p/dS?.Tbjb2E/1mX<eeWghoetsFg_!8\hVa==/BRk&$!s#r7l.`*`fG.c&hEW(G1egFu0Agn(Uo=c'TZhS"`tF_h6I>QRnu,W>Ap+0*lZ6>P/?C=#Kk>pfEI%X5o!bF40@(o?U'<]Eiua1#T-.-6qJtTfI(.G1oN.lKY+4/`*/<nC_FrBr@t'B!tXgNRb)RL+TCgW3<iX5O9#P?a!)IFFI8oG32\=#jga2H_&2^^0Kf,.OsLu_#eO08A/mmJDDtTcmr4t6O1`Dr,Q`30k4J%!o:K3@7,[V(bXXTZYYa2Cd1qoBXD(l2cQ3/<j,7X5ml5p#+n>6f$6'lUmhZt>sFRC2D);h+q6STAmNibWJLoa!_K+fl3/2UYVS=VJ+Mu+BpgT4#ns,rG4")XqHRFhp?e]lW)6<M/i!6i]r"JcHp"^;CaL.d(qc=(QVd0f"!@6_597/@.grp/FPoFQ-+&Hk[Kh<ZWObTpod[MGb+)FWKf>msB7&pCcn]iARI''Q3&WZn4`cfmXRYX$=L#_*oT6]L\$1K[YD^6ieQ7a0Sj]d/M^g5G<=fIGJCn-7*ka$`dtNA96I:UC5Sul1`\4p@]Ls&$eZG>,T=j\`^pYF(5psRDI[ZIJUoPB8(M_b:GbMQoKNEpNmMNdcINqeCi3'cEM:qF.X06_nLu%gkBg5VlBSp+B1fTm,9!@mC$E:N.o^gaK:4o/&_:c/cq+m2[#j^;N_DJlWf4=p-!+I-6h@omP!WV_Y"`IN?:CP*<`.u@#A9nDKO*4\G2pT\?kZ'Dt?,HQ7f!)_^L]dn6@h4u8d7pi9\@8;-o?'k"lKf\AD;4odYK88Q^$"H$>rOH)/`G1DGh5_0eMt&PpsLK>qs%#:d;7WAZfKR;rBm5'b+%b!VCn!a[@b*Y1e"S\)QM"QV-!NdrV>Ws'[ojRr<li1<loMopq>5.dVU]'X/_ssN#`kAmA/umL!VA/'h#6Ik/uc^1EO5Ek,(eJ=3@$n19'-D]1fkFPi7&q%$9QZs+c](m.qMB*W<LO;^VmoE[;onAu'qSYrV;=B=>iUUHSFKA3ttm5!=55Q\.qs8;OBp(ii!qKaVIY^A]oF]ER6$H=l(;gJ?HbR]':Z$q1FFKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`fq"Z!rjOlg~>endstream
|
||||
endobj
|
||||
6 0 obj
|
||||
<<
|
||||
/Contents 13 0 R /MediaBox [ 0 0 612 792 ] /Parent 11 0 R /Resources <<
|
||||
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ] /XObject <<
|
||||
/FormXob.67f2b803142796cfcc78829acfff7782 5 0 R
|
||||
>>
|
||||
>> /Rotate 0 /Trans <<
|
||||
|
||||
>>
|
||||
/Type /Page
|
||||
>>
|
||||
endobj
|
||||
7 0 obj
|
||||
<<
|
||||
/BitsPerComponent 8 /ColorSpace /DeviceRGB /Filter [ /ASCII85Decode /FlateDecode ] /Height 80 /Length 3671 /Subtype /Image
|
||||
/Type /XObject /Width 400
|
||||
>>
|
||||
stream
|
||||
Gb"/kH&NHV(WW/Z*isSg&d2e95Yqb:,+b_Ylbt"c&J%p\W#GlYi-cjh1$F4VW);.c'TXPH7"cs*'CeNo0aOLC"P*Zn'GO!upO2q-m[9Y1pX_cEoNV.hZ!I=.X/ihg[pN.eCb@(q63n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f<&gLu-H0\W/S)K\UiU/d1e=((AE1\ViZgoP7G`;;^96"eA^VhA0L.[BPc_E\[V_jF2]4eaSpjj$D"&Bno1d#['rMp*ip0m[:kfFCp?cGGD5F+!`f-%F.hd'N;+G=82r*?QTUB\d3]4;&O$@@]/UdP:"a?Mun%X(hIdfY%c!Gar_>X+@1HlIml`F@TKbp&ZMDsE$keh8Gd3hgr.uuh/jX)6'!qj^&c9!\hln?+ERl7S6Q>-NjMlpa9'WJ6Y0"C)9EqnU6a<@k<:/9+6qo^@Z'\I'[FS"YZTL.@L2sJd]G1f=ahJl&2l`Ks-M:S_+:'iN)f]\_,l;^8p>qsj0Lbs?q^q550MlTodIT_d4`uhTtM2WG=S:0CRJ?g#U89K(O`^f6E_&QM!\8c8?YcYWG^A,Rga7Gib>7NblcZ\TL@>T?RFh4gP,RLuW%NY0c='mP/s6[2ZP"RWEO$.%PqO$8NHF;:(g19^%:Vd53pXdDSgjf-D?!)SSY:tXbbWl,ljiaKo6&,Tk^%ZEq<rWC2djpdFNmk>T*#!;VcpRKUM_Ah@JTSpQ_,km?"fI5ldt/$XrDiRJ>7IaG`llTEkqIRJqXuLseKJ0E@^B[c'G&YC.*Td\lQ;0M&l<>n.Lhop@hJHBr`p<Goua16/n;ob.>5F)Zdo(A@eK#h]C[X$;ni/uM_oq\m:G*7H0QjW]3@5ik9%GVKE;f&Uf!n]DI`Nb%2C3boPu^,\d9$\T7(8@A:HcW%-c/0@u<e?e^USp7r<*.WB9Qj*&d0_/#)>2Bsh8UF<RaQg/[=u(e[]i3HG8U$[j'W<'t#[TmqC\O=RK\k4gLjcBXSgP'657?9`D%/7'<t=0\%EfUaok*dBq#D;/+Kc<ku7gob>(_H9.A]-D0Nm^Ygqir3#\O:*\f:`5QV2)9W.6-PkGX:u*s.q82:[bLF*!YSl>fWgl`93^mI>>?W-=Q[qRY)bl6lGcHZFNr),2$%FM_ALH%]g?+ZMc<[[9]ZgI@@Xj1?$uZ`fQ@E=T_=^Y)Iq>Z]n3T+,ESq+[<(gAZ#oB@"U39sQJhH7qWUVC,rjBb5Bpf01/"POnP#E1HLYF]r-FX(;KH&I":;elcp?*VMmI:t9XJ+lKojSD4*?HTLsA.bG26.GQ$>[hlK'QDo^(hmQ-cTH%506+oa4EtR1#lVq>$DdUYh/>J)/5O7A3XUoj?\?S"2`7HXiK0'jc'n-K65=5Xbo07YG+,DnJ#k)B0'A/6f!>6\]:+"l=`SG$XA)CADn0K'_K4f/f=5]6+1l9?W_X6Ot?rDnk[F$>,)cO;]%-S-9:BXpp/7pgGNTNfPU?P,hXj.lFe))E5rE0u.QQXBgC'[=5fD-+Dd7D_BiCOsStaKInr&6L*#iQ7hibR@-51:cbp\1q]mqe13@abo2#F%iXN!ou3+QMh*ChlU('RV`>Se52@/A>k8Q@LYVs5!&,>nq4C`OI,m?KN"dj,lR]4^*`J4t3RN0'e>.R)(f4&I7-<086hRGl]?\CGX4ZLuPqD?mGbY492PJuG5NhO5Rl'E?jVGU6ID)(X#%WK)7j1.B[t1a9Ju#GK#qImB754Z:n*t7Ur]Z,U]e<VdfHLtQ-nqGDhosqd,=eUe.n.A!MBr':nf?;3P9TgoKmg#m"o/ECONol,Ita.<KBmQk4*.;]m50gE0nc>k*9ufGe;5djX]Ln4+n3J_qX-GkTR1n9@0\q1VH9&8FY9fC.nhd]SpA>*.D5WHRi4b(7#_j-Wf>90+Nlk8XG?8Zml*$eFnI4mV<54R5pU/km!].$fd-Iknlq@5,Xe$Tq95^0d<lAV<+_t?GZbWe?PK(;2'+J=hKZqF"CV9=i:=Sc3pXBNm0iUoS9^uD((5S3VosJdC=XqK(Y2"k`E5Uq'nDYoh05K4psDTXB`"b1or?HP0("#t;7\'`7fWI:CgYu@0,G<mOUj\+!#:VG'\hb3LlHB]mfcAP9ZI;M,2&bnaX]6X/Y7e('"4F+\QPhZp1:%8f78Z'U.)Ue6KD?ha_fc%12pV.ZP#..XGC/#0BU7nKDib`:Hn"\?[&('o]Qm.c(BLBHqjocLXHf@LM`\>@eFKu9Kg=%YequpeAKtGp$Y/ZWq<GeX&l?&`O&_;(00M@OlMMsp8pnZ(i0jKoBoC=e\>Nh>cB:jR9h2CeD)rjW#<;*rq4m'&E21"B#_8-[n2C1%.R]OKZKFC,\A?;GZg/0Y7QnAmMs\'7ippJ^\Xso)'/EhB"`5cHhZ>EZWOn-3.t*;K+PqCj$o$J&L5tQQ=@Q((N`qd]u'*NP.O+%`e+d_QG%XgkgA[eF6Dhl!:9<aH2$USNm2LWq28%<VYR)jaXbV4YCJ4u\jDXW7Cc+bW^I:L/(3_5/$Gmk<L%t/D89:Y9U<>n!ZO%2OGm.G+*H6L3F-L($4-&[X?+S`,)XG+'c8f#F"j,#4M.6<MY4!T\h9(FHlkcV,>FdOF]l)Y>rspUrd+UDb:`D!)fGlVc8C*chooP,GPs!_V7E]#$=\U/qWTG4Pfm%09%<@9,->21E?O4?&pP2@H$`4$?gM>J.^gGA7I8>OOjhu9o]#%SpYD:!d-.[JU5C>G.uT">4k:L+m`uN(or>=//s',t<F).:*dk3lKC#=$qW5D':K;=$Vh$e>G-.OlmLurLLn0%0^[#WL$M5f>V7B:m$9_l<#pr?m_rNDl<j,-Cn?O7'?6LH#Ilm/to:\)&a/]7'o(b@./-_E+cQ^)a4*],55ON>nbM;@Ec#YLh)mC]GG;L?q"@ifc)?NL)=1E(%%]V"'[:(Jp]+fX=<H2:\81X=InR?-T/csEXCV7l>pXRL(KCof)C>7a(8CDp=I/^[HE.V#Z]?\'*R<,cj#3Qd'QlFa+LO?d-;J@`k]u&gL%D;=rZV;%XY.6PuMmCm6;Dc%f8>TCO-`\Q1J;CGJ(.F>s_rU",SE[+39$;uD4Q!l$]cFl9np^it-Z]/KiBJ2.rd<i1"5=SRESH6hk"I1b&.Z[f2i1jldA*7J@TMK#qXgf3]<7t,gRFY%(IWDRW[lrsp-a6$p3oYdYf(Bd^OH#r>$B]'[tKFhF0C=aRdt[f,Y&iJ]18[YX+V2TDbj8FlguYXiND`1&gV9j[X(r2L6iXSoZBYBjd4#Tfh\E%5A]:QeC^]5U+TaDlRI4g@n3(]?$0/`*ag3C]s<bYFJo\=W[`.=4S7639@Ps.ou\;sq=0D>YKFND8ubs#ku->#@_ZT/BZ#nhJEtc$fKAnu*.>2c7>0%$]DfZY`<r1%=hp-6L:goFqV(ALl`"4(F?bVarqV"-(gH8)Y?-O\1!d^"N#`kAb+5=sRHml8%4?f?63n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUEtn)Jn'&1>[~>endstream
|
||||
endobj
|
||||
8 0 obj
|
||||
<<
|
||||
/Contents 14 0 R /MediaBox [ 0 0 612 792 ] /Parent 11 0 R /Resources <<
|
||||
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ] /XObject <<
|
||||
/FormXob.7ce3f428fed09445afad362830e52447 7 0 R
|
||||
>>
|
||||
>> /Rotate 0 /Trans <<
|
||||
|
||||
>>
|
||||
/Type /Page
|
||||
>>
|
||||
endobj
|
||||
9 0 obj
|
||||
<<
|
||||
/PageMode /UseNone /Pages 11 0 R /Type /Catalog
|
||||
>>
|
||||
endobj
|
||||
10 0 obj
|
||||
<<
|
||||
/Author (anonymous) /CreationDate (D:20260126185515+01'00') /Creator (ReportLab PDF Library - www.reportlab.com) /Keywords () /ModDate (D:20260126185515+01'00') /Producer (ReportLab PDF Library - www.reportlab.com)
|
||||
/Subject (unspecified) /Title (untitled) /Trapped /False
|
||||
>>
|
||||
endobj
|
||||
11 0 obj
|
||||
<<
|
||||
/Count 3 /Kids [ 4 0 R 6 0 R 8 0 R ] /Type /Pages
|
||||
>>
|
||||
endobj
|
||||
12 0 obj
|
||||
<<
|
||||
/Filter [ /ASCII85Decode /FlateDecode ] /Length 302
|
||||
>>
|
||||
stream
|
||||
Gas2D0i,\@'SL]1MAmuWBq2]41U`B=+JA;@)ETUJW/a5S$%>'U)FQlACq\foqfM9dic+ZLN<Q`tbE@K*:^spj"Oo"1`W"9\N8S]7/.WgR/fq$*$ITZl?0A3Yd+#RVYd`S"!VHM:q3ue\ZE&.5ico>/#%%PKVtVn!b+n6KWeM,?U:f@u6(=k$)>9=A;GQ#t3m&eV#g&$:bL-jnalu?/Fi#S%7?Zn?-:G9#d\O:D4D7XQ`j*RVq8@Qm.FMjt9rX$+<uAFWrR=.*pU4ORU>6iZ0lp3O3um&1LmEd6.tN*K;n6j'~>endstream
|
||||
endobj
|
||||
13 0 obj
|
||||
<<
|
||||
/Filter [ /ASCII85Decode /FlateDecode ] /Length 281
|
||||
>>
|
||||
stream
|
||||
Gas2DYti1j&;GBn`BOD:C$`F9ZVnjI&kF'G*Fh#tN?'!3]KRM,ciKk&h=0aE:V*="Q_Ne&,OhA1lR406EhL)):sXAaA0[ug"g)PsmBSG*k#J$))")C&+kr+KmIL<Brl.":L6#Q;:T1n?*25E!Zk"i,4uuBV3G4oRN56iFD+G.*U'<hlkt*7N8pVC@\#B7T'\f?qTfO:fq24F=Moh9cYOO9_Ug3_JW1$`&3Et?9G$Rf%HgIe&37c9!:H9)*A"58?9%Ib;S.e4E4@\m25^i]720%7~>endstream
|
||||
endobj
|
||||
14 0 obj
|
||||
<<
|
||||
/Filter [ /ASCII85Decode /FlateDecode ] /Length 249
|
||||
>>
|
||||
stream
|
||||
Gas2B8INBh&;BTK(%8)Q2+a"?*aMq\4K.?![FW8"e$p+akF8n$UkfH'`dI4a"80L!>ZbC60Zk4LLGE6]&5Z#qMYu/6Ns)3ldF]OCoN(cR,K(-$>Bb@Hb$Fm@B;e+Uh$?f>L6HTg25p.\@EBp=GIr"0+>.bL"Ab!5e$0H>2u,XrGS3n+\I^LXNi]kl12d&'Y,la0?'!jr\BDiS++DQrec,bZT6(6I/"hnM&*R'u?RM762ns?o2j@QC[f~>endstream
|
||||
endobj
|
||||
xref
|
||||
0 15
|
||||
0000000000 65535 f
|
||||
0000000073 00000 n
|
||||
0000000104 00000 n
|
||||
0000000211 00000 n
|
||||
0000003775 00000 n
|
||||
0000004033 00000 n
|
||||
0000007969 00000 n
|
||||
0000008227 00000 n
|
||||
0000012088 00000 n
|
||||
0000012346 00000 n
|
||||
0000012415 00000 n
|
||||
0000012712 00000 n
|
||||
0000012784 00000 n
|
||||
0000013177 00000 n
|
||||
0000013549 00000 n
|
||||
trailer
|
||||
<<
|
||||
/ID
|
||||
[<8efaabb9b9953607755769fba673a5bf><8efaabb9b9953607755769fba673a5bf>]
|
||||
% ReportLab generated PDF document -- digest (http://www.reportlab.com)
|
||||
|
||||
/Info 10 0 R
|
||||
/Root 9 0 R
|
||||
/Size 15
|
||||
>>
|
||||
startxref
|
||||
13889
|
||||
%%EOF
|
||||
@@ -0,0 +1,88 @@
|
||||
%PDF-1.3
|
||||
%東京 ReportLab Generated PDF document http://www.reportlab.com
|
||||
1 0 obj
|
||||
<<
|
||||
/F1 2 0 R
|
||||
>>
|
||||
endobj
|
||||
2 0 obj
|
||||
<<
|
||||
/BaseFont /Helvetica /Encoding /WinAnsiEncoding /Name /F1 /Subtype /Type1 /Type /Font
|
||||
>>
|
||||
endobj
|
||||
3 0 obj
|
||||
<<
|
||||
/BitsPerComponent 8 /ColorSpace /DeviceRGB /Filter [ /ASCII85Decode /FlateDecode ] /Height 80 /Length 4030 /Subtype /Image
|
||||
/Type /XObject /Width 400
|
||||
>>
|
||||
stream
|
||||
Gb"/jGApR4(<3O%]`e\8(`3O5&gh'9UL4Y/2^+lN@RPf5J1sk+NQ;)EK[=iEKil-Qc;6BQ-:LGDM9+$J',f.nAHF$>+;)0QS%B_Sc9&LGT7Z-dn+T=!BA[iT=_mDCmsXW7DDr_l&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j4;C$M\fk4SCf9@^_*csp?0E:tA:NU]#jiWj*eoUTRh)0!!`5]fOL5)!H>oFG9DVR2p+lUH`IrqKY)BXD"&`V6FB2;HMc(7';GJ3Ri.fjIH*^/fZ0Or*2Il>Q?21r^]?[Q:^FZY!?_$=YY:S0iN>)NU7,O;iAF)l:J:7R.&b*4>RZXupbn&1%rQH_7Vb5!5MER63=/j;H?_mC;n]Y$@HQVU)1)MO]*WmC]s?Bm'Eo$F't5&H1H?C?\PZOT*=k"OUBF^Z7('_KU*cW%)S*WMHX>B\o<I2:$`SBCZ%`^?q2(GB+]eubDeR*F:Vn)#4P":#10VP[s<B,;6r>eYSG,9p^:L_3O7EcSHah6r%e]`O=YOgf8dp9lDfH=\S3c8r1HgU==+3,mg#Rl>?_pYUI1'87(cTWVY:DUqL#0^"?4&$]I&kNC0`5JLC0C)C?hEoh,WmelnPO$)t=.f&emDnSAh"(gjeWnZaC>tjK_q=<Y;2^p2tgSS*;Q.a5>kWnKp@"uqSN>jhKjj-*a*6R/cmlaT]DPqQi#i_sfOu^V];l<C(qWb+)+X,K]55F9'T7-DN5.u%#%NNom8J@WaU$Ys4EcZ<p'k6@=H1U1Nf\"4dJ%Sa[;MZc\a,_<liPGacdte(N4#Ld&J7G4#qWW.gemV!7e*Ykse!lmlI6&MpTjGEYW""=AdA+bUmFtWqp@60F94#=0o#op<o8V#IIK09?Vu_>9H0E/'"u"N,<U8_fPGUCUB[J#'4)`ugV+[/l@hgImB]$Q&++O4II3)@-X=:iG5aNlrihrDt1>!G@S%i^qIJ7$3^\6AsEbQnB1qh&Ubj<rb+7oR9`51Wc:HuhG`mCl"YPN?sfuR?9TKbo,*Xsp8_GH7iUV'<j2Q"^R:?R!:`1LALo[6Bt.TJfM[:mr31c/1%Y[kk=hS"9r+/DM.<0XIVdF$A<$KL.*`4/*c#08_`F9A=5:/6g][Wq=O\\1b/3Y3fE$[VL2A^Dsg+d,P8$hWM:-_?FRhKnLi6AL;36.CAZjVR[)*:k&[VG3Q>UVt*h^pT^rHWE2?D;2KcRo]!jbUdu=J*Y^iO7NMpK&+/M?E#p8P[:&KcCI&WT>lj0hn!r'Ddu;@XC[FI)\EY_V0_d]7q%$Vah7c\%+%Y5*O#<]LtTjQE1fFkLQENDq6=GM6pd`rMIpbfSH&W.T3_Ol$<b!Gm_&PqlR5%C@JK*Ol!h2Imp7Ot.+d%:%3%3]MgkH[#Hbj+HhPP+H1'Iu;KCj>&_r2rYl%'!QJ.^n(hm(#X5AF,*LSQi,0+l+X_QCd-sXH3[JDTElP16rE1kDGP]cKR_8u#V]Y"6aMOg*%"icN@-gSBDCX=S#a-tPZm-JQ<M>nqsR%UpnUK?#%8%U]B4C%d;CZj!6'e<<Q+hZci=8cPWZ5+GD%id30L5h:gr7\PodKJi:2tP_K@\Qq+EV)a?Ul]h5_1DjI[kCso9J35<SVlOcqg71^,=fU%5!E:*Yls*-m+ARh)lulg3U$-Nba:,pm+SkJTsi93ruC,0)`CY;VB`@`"Ms*Z\d)j&;boQ1M6h3^7dju]MOg*%KdG1:BYpEDMN0Qp=1H1W:%UjRoYunt=j%eq(Yulu6#ZJULEF)i>=uH5kuE5#MQ?sdq?&m6=[kl8Tc=t$9jhO62tP_K@\QrVip+eZoCIBVVI.)e.%EMOIc+6PiK4ajcW;'k,f-)WZWXVHl1GC1G[.CW]@LAEm#oop\)2X5*2\@n4)j*X1-$m:bbYF=S8$HLkmmrTSX5auF,e"0nU1VTA)3IC$D8-]"Dr>:d49"#,PSWbhql]_\g6^db0"bZp1dtL,AY,Hrd`/-m-ruOhB-14LQ;od4K*.0TNLGYVbWfTAdFC7kkt8JqJt7[c^Ql>:b?jjnDUs$l_[IMNbHS?"ibH+Q6-&jp=Nm3cV`0>dPSYcp:A>VG$`7Io@6oL.1Yru`9tj;1MbRC8OuD!T$h`FdRA1Y^%4"c]QM>g?3P;LgT"R'VH'Wq6-8?<USYnh?<PGk\JN:FmgdODqo^Y-[-q!`!^tV."8sD%$a=W62-nVR5dA`f_`,bBeeu3ao@Bu0gUBEIr:BIZ;o%Z0&e^p5.nho"?_Kdin'6=Th05;oSNV>N<>b"^YjQ;nmbG@*@ga"'Fml$&HKSjO@C'C@dp'!_Ffa>t?@e@l=1ULa\_XlA]C"jJ[EOb[EU<Ad0TKc?#SM"3X(Lm^XP:Ai-*%LAh7FIFeZ)VBqd3-*?CmNmaQd@AW)n[b?#",SQm'MSet=LGn*8H(EmU"ap$8frb$7:h(cmkPT'j1f=')P0OfDf-J!fq>LM]CT:sc(6S,=-4)`A+!]_LKEDDRhbe129S\o$XGLlIB_@GSM;0aiBoeY#3\$rnT",rr#-L8WThrVH3Wd>f5/m!I81VBY=an%\pL-#\@;=L#`iU,5`YFD7-fMIo%:V-`E0oiVNt"U>:mo'NpD2RN%p)fKE=Wh?#X9URXg?^k0Ij3g**Ork^G>.)NP0^Zo`HhZs,@E=NRrX<D_QiVi,Ql*A5m(A3^.6?$s:Tscqo1s(3kg6"-]oontmG$5h'tiM,?M3U6bKrXpDQZ+8OY)5\YPQ1RA1]df+(N?OL"VhJ@gqIkI.E-;o/3Qt1BZ.-^fcFk]\"'SkiU-Zp$1)VkDu]4'.6Q)Rpf>e7Rl\9C@L/t\;Z<&1+XN&%j.rRWDZ,P`5RWN'o-KfFuX0F4HJ=HdaGcm]mfpk]J#M4P%(H_.XIrZ=?Cg4eurF6+-eE^<^5E+044/<]H!b,^3/b-4Pk'SYL'H2BJX;H*1(;>.@2s+l4^Ld[GX<"m,,S8jnVW1p25U-)leT"(Rd*85eRMpFgl;HQ5B?dNZ>%sGi@`*P8u]+OF+6o8`@u[s,9A[)598-5To.NR<lP-If-_B`3<WT]66n!$kEk=@IN'deV@j'G*jECo*=n-CGGPSQS2^c0JU>Hgri>q[;4C6)9l.D<V/o>Yr;9tI4q7F2Vk[EZD9m8%k8qS#MUgZFAT/+X&c>ZakEt-!tIC@gq7p=,@:&"fuR?9TKhl$^"^,@C[&CBoS#=R9q$`uOH:#JSJ9<W:p15N\sY?eMAc-TfVUQCf[/aU7Q<+W&cX!Gg5QIU/.fucFm:(bne,@%k0<G*A&jUu)&=dF=S^2O(/T;7N!1"hV*7RChA=/aUheSb/jG#CKc/_f;X(jRo.*8Mg=Ih\Oh@5uQu:VmJml*(fb1#VW`5t>P:&Gm=.MEs`m+]EUC$a"&A7V[4&1(O+/U5tc%5l8bfgduMA7WcZLS/+L8KGT=PZX]or?B?!uj1:N/iq<5oqK2W)9>[j2YeFBE.YV?a;6JT9Q4NVaHp2:S]C*Y[u"D`JYPMk(OUXco6M[L(>@Y58HVo85]"55<n&T0HJOk!N-,QmPlFjWD]R'acb;QGO![lac[s)?XtR.?N'ae(!#%[/$O0^<q#:-:!!5#^Yc+q1Gi1@C=S]=(`^A8mFp['?G60sRthIolIN'VAu5FhcfZf/QG"1R`Q25(?i\KC4#^p(-nO-j%L*Xa(C,)gChDno6`*pRMWEi/GSmFTTK>IG+o_]R.!G(9muq-BFa8ETq]Ipg#U)AK2f>//o4/S?>UdM"B);/a.)]9Kd\TSI[X3Z=U2f//"o0>`-h5:!aHeD^"pG1@4>4BtrUnbQ\p&f=niu3sje\cKZtRhg)ldr?b(YV+-RL1_Dd!GjKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U/qprr=Mu.$a~>endstream
|
||||
endobj
|
||||
4 0 obj
|
||||
<<
|
||||
/BitsPerComponent 8 /ColorSpace /DeviceRGB /Filter [ /ASCII85Decode /FlateDecode ] /Height 80 /Length 4649 /Subtype /Image
|
||||
/Type /XObject /Width 400
|
||||
>>
|
||||
stream
|
||||
Gb"/jGB=Qg)obtD(hlkN"f0(5:+-aT<2CgqGS',YK-J92</AC'U_i9g+@S\\U&raT$(ug!$4%UW-.5X!OX),AnUAfpn!Ei.ZhN>CmL:6LTAL_NG:0`'1[i!JbWE/;*Y0EI&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j4Eeer2aX7P8Gl-m;mrV!9$er./Dr&!ITgFGgA]g5nB?j\gC<`5,n"5+/EDb62SNDgA(+`SG0AE<rQgcQJG2U/e+R8qjL(>AddU%7a,tE0=*^(Ec[;Xqd5T\EOW]FtK0RnAjQS4C.$PtF;o^lOY2JjA(q!>?5rnj<G\:)&,dhUp&iW]f.:o[KoXBDknr:%Vf^;G^:].Qgi,t!Cq7>hqh;qnpMFKAOW-;:r<^AH9g>e+lD6ps06kbI-F(^N'<gi-Gcfoqm`D<`e/XBDl/[X1QKi:_Nlme*"2?`R7BfZQ0Yn][CW7>_eACt5OcbEjk(IV6iiD9J4s/kRp`OHAta@uhcDp$1sUc^m;JVdmc-<LpXGp$)c(Hk;,Z7Z;:iTQi5(W(^OWj5YQ"X&GpV4SR^\.hEaCrqG<">P%b#odXg*ftJs\9L;@@2JoU%\UmIb7]_:X"BnAg8PVo7"#l=q;QoL`\om<c?*C)#T0=:[KaS]?>+g+\Q7Q1-1hhO`F6CiVjuIe^m/?\904&j@l.#kH4Fk1D;,Pn,s$FCkgKq>QM?oqqM4!SS5Q<QQX'N=qdO.h^m&2`sM\[n]^V[k)d\U9,\%pa.@q.TDm$L"eIRL!nb*BjG)?7Bqo-V?%d\TU3:EfTg^\mXEJ,E_-&<#c6bEkHfgiJN=njqoeR5#87\R3+#(G8t>*fUYEeuW#*!X5lBcX+"oeKmkS@.(h*0mT1nrUS,bYIsE5U03`ScpJ<e8mf>^IB)[\rqY`>H2da;>5Fp[LW%H??G5X>9I<PAY[@K\1i1gkRchBYhS[+bJ,aqhT)A3+5BlPN)'4Kg>8jeZbo1@@Rl5+ugph@\]QnSZaRRJ1c_/-=odWtJ379>AFMnp+G4!`KBR8dD=K*Pa.$qbpOj^:tQl,H?<^#prXH0"q[r1$MSsu`#-;Bq^d`.=is1np^^aDK96L*.(Hg9*0T*>V9Q`\n^`L]5>i\sQ)AOEM\?DBt!8#7Y0SiiDsB23^-W`?+JZ!P-=ienX,kdX6M.]EUBN#=ETZtP"4E0#kk.nX&MZXupQJK6dnON]"CP^nh:H5s_#io8rs2G=?r4"L]CJn3h!L8UnMn3&`J&s32P.9WsPPW!4%+F!3AJ[djYeuWV-Z>A4"8En[*=5Y$_-RU/bi:`*I1]ICNmogd^&>KHo<_nI&aUW0Z4F*r.YHH(N0/])H@AVt,Qj$t>c%"]+(Grh2@hqR\Kp>)Z"qC's<2ica_Zur<m^u*Y/Q8MTPR<N]nt9$(DsPuVc&uYY%\d#Y-N4c2<Xd/X\=C<^SfiC5&Xk4B%BU5u_1MuTN[G%fFL=0<a#bps%YP9MDr1EZ\)4&m]`M"L$&W*k$l5Y3iR$k4ldeYQkic^(a@KBro!2iME/?\=G3i$/_H*tt(cQ?&U`e+$N@567Stob?D:u4k4BLb^V?V:WLl&4sV)/U+,U1BRd9b&3,)e"jWF"OBU.>-Q2/AL<_pP5LNT<5uc'*A?hCXJ.kFHh"?b\4MT7?jNW:Z<';^CI[++A`7OE8^;3KaGt`tU2'.D<$$(8lJ$qXeKD*.IYNhqsqO(qjtQ7I&aVcqsCYCu`NpZ:0D]`fV90Y<_!ZI:XrsMu4G<QsOrnk)*8Si:<@U^<s54-76lfcp@@u'Ae03>pR/S`Z+]D^@_i>hXTXH<?g3fN.t`mHFrIk+[^u,Nj?A=n'Rm8Z?>Qg<A$#ni\E8Ed[UQk3^PU.?M3aB)jcO&2:>*\=T.d1+*ZF@"?A5D+<XGJ21+oBV+^@Ul)1.3B,E[O-k`eb[<f-3'8ScY12"q)NH?`MY$JYo98TG?o]]l2K>E8\PZb2+R`274i=_&+Z-TjqgJjbPoZE^@ah=VW647kCR58IohRMC(*CR(BT4n*g4pe*Q*FX(Ze.='Up?^1I6=]+CR-!_%#,)!P&(,gr)C9gt'd@U<[_Mh<bGWalRRZ:i#nmA)7@_#-gU80l>7[.BXW9QFj@HU`+`>?^il-h`Cm[n-7Qu"^R/N<pp-\T5XtaG+fY:'fp1.8-b7N1^\)2X5)8a8-&-1WqRO?!W,`XY#S-!.OM%I+5h25bRQ9Y.]eP[9P9!<'"`J%WLB$Hd$-E,<4N*a'd,.Y1#h7D;b:aKfuJfOZ2&A>q)/q=rVY'\h65$\b8K9Z?3pKR74:;W&Vrb0'RUnf8X,*n!Vo@(0TeZW?;S5&\%q=Edol##.]0ta!D>-S?B@9uVpR#`p(A3D=o2%[df2_8e$4'f1)NRB<l#/epT=HLX<p$1'cRu'Req!dUQ^M`WY%C7F7G4"#B&i77,rq+Z8\EqkQUW6mqn`1?2:<99X*Ib9nLECu$5p=q2dRJff._W-+(3b(Yrlpuq[pdum$V%>TH'-m?jdWZp&qR5i[E?3(7'GO%!UQIuh95N^kDD#a!lS/(+69W4/mZ%*VQC%uHIj\7D6iFIm5:M9YL]ma?a!d!FZRM2L8hJQ&\WdViAY@$CLtG[U/u!RSi'E`c5,!URlB=%P(1u[;9l7Q6M'9A^A>u+KnAe0>frXsk/mMom5)EOm'BeFBCOfX;l@XR`#.>7PVnNg$Ar0C2iBc2!hXl2M:?(bVG3Z?oZE^@ah:&r%'`jC6A5c$G<'9m%\d#1i<%XtiOYApFZ#ngUWnJtG/]%:$X.K=GW,cd(,X]VC#UfU)`BPA]i2*s(/L,')$%Fg"HB/6o9V+;QA(poeWD(H%'PK'i*'^M$-GEjj5ZsajEKE@7)e`B&V9IbT7,k5*08(&Z-#ERJlX!nJ6-,qOtU0+)>0FG+$Y3Z0!U5:(0dnE2>ldh:HuO;nY0Pm-Nb)*J,HR6XB5,?i``NJa]m"YMA3m7nYoRq4LCiWU8$(:YI&.rTieR/pt(60)sl<&(qm64bGjc,Wid`T\j#uS,;!3t#ZWX@Qb]H6m+0<mq"pT?/gh,$$MM]52_Qf@HL!0M.BgGYRaS8&f<<*5L9LAQhWCm'9YO$4=E`:QN3t-8WYjUQ_:E*b:=22WPC.B]jq;$$D\t?-7XMIQbD+4-gUCs0@Q;HpGa*a;+>:1;qsHNtRn/+a^TqR>@.Xea;s?o];:@&[UK4L#Bgp,F,RsE=He%6J^-n8]*f2^ig*%<Ho%<Bl^t<YGaN-n_khWk[Q9Js,*5hZBeUD3LcIOkYKViFO>WUU9]mE:;]ttcMZNm[=\K]o`.A)aeHbb.4k%p-knF1D'??PP_$'uAW<n&=urVQ?PaH<5kR52PVqQ%ASi>.YfGi'1*4F,A`4oAd^jGb*;,,KJMg7lSZNi\i-]QnS9fD@keR"@(a"19:*>e>/RJdS>U2U)kn??q^kNZ$^Bf?AOeYG1M#F69N)YKHC81t4$<='G]b*BVjAL@1)g&>WXcmq%"$FN(@d[u(<&)jh7^9q42j;/'(@*qZ9#1_hV7kg;bg1;[eiWMc>NH^cj+,)L?e-tC8Ul?"$BW,("fP"k2kTgOS\'#P]UR$afb6UO5'fV1fm!:>pa-mFu0fN>Vk8V,EUiAp`*k=7lnd6j#G?@gXj^]4:[lhRS7^\h!<=Ohc'UIUA;HcM*b-SH\Up<+TaZX2<A99=J]8a]hLl'35R1/-l(io8r/DFn:Ul4p7&\[%C"A]pCUpdeZ(I(:I`"K>JrHeBK!>nJAaY?kLL/gnU,#h^q""dFhRG&6H/a5UeX7Z<FFkZgN[Qi[]b[qUYmnY9pRZKapTW57tpDaub.C],WPGf!!siWZ[$Gc?(sK1T!bnhJ6m\oc&$UX6PlA0M$=gp2k0U'g^N_FXLaSPN%91<W='3St=UOZ_!6o;:`G7>p6gFeM-UOA/J@Q7cIs89qr*N`gtc.gV7W@Pgu35H&1[il-gG6ps9s12"o1k*p:dX^3kuci>4)9#`+:[3-;KGd.!54*Cm-Y<5R+fnq"UN/<B'K;7<YI,ljoR\iis3%@XTHKF\iIdi7K^8P2@-It.qj05blIf9,65(+?-)OUrW];+]CXb/IHrbtO&cAE>ejI8lFT$382/`"$_QOh#2/DLjqoXT1J5gUR^:Q73Z/%*Q8G$9Be]Pl[k/.(FEb(9d)@N&6Z]ZhSo`8H7)_IDWLQ('dTVL5SII6X+!=b>6UY]Ahta^E[MKH*pf9IX>_4DG/tD:u3@Q=+'LrO%cbH8T["^qG*h2K%:eK3(85o6G^0<BC>e=,qT0_iZI$F6Ci^qWb,K`6fP]$4[.MF'Y6B(&,:Gh,TCP2$t[aqq^Lo&44Hf_5)pDh0RQRF/\'r*qi?.M@`+%+QqN0=0@MG=&Q81PV9AI^:8FXigm1m+bV6r>dtn(*3_sE%hF_WLlbEq6:+#iY$HCPCI\XR.7d!#(c,btV+R!a:h@tE4Z#"&=0Gs$*@i:d&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+bUCn#U+j463n`f&4-XGKFgHU+lr@e*ukCHmf~>endstream
|
||||
endobj
|
||||
5 0 obj
|
||||
<<
|
||||
/Contents 9 0 R /MediaBox [ 0 0 612 792 ] /Parent 8 0 R /Resources <<
|
||||
/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ] /XObject <<
|
||||
/FormXob.41b05a9cf8679f0fe6e7c30c9462b767 3 0 R /FormXob.94284ebb61fac7951963d5746d1b193a 4 0 R
|
||||
>>
|
||||
>> /Rotate 0 /Trans <<
|
||||
|
||||
>>
|
||||
/Type /Page
|
||||
>>
|
||||
endobj
|
||||
6 0 obj
|
||||
<<
|
||||
/PageMode /UseNone /Pages 8 0 R /Type /Catalog
|
||||
>>
|
||||
endobj
|
||||
7 0 obj
|
||||
<<
|
||||
/Author (anonymous) /CreationDate (D:20260126172022+01'00') /Creator (ReportLab PDF Library - www.reportlab.com) /Keywords () /ModDate (D:20260126172022+01'00') /Producer (ReportLab PDF Library - www.reportlab.com)
|
||||
/Subject (unspecified) /Title (untitled) /Trapped /False
|
||||
>>
|
||||
endobj
|
||||
8 0 obj
|
||||
<<
|
||||
/Count 1 /Kids [ 5 0 R ] /Type /Pages
|
||||
>>
|
||||
endobj
|
||||
9 0 obj
|
||||
<<
|
||||
/Filter [ /ASCII85Decode /FlateDecode ] /Length 300
|
||||
>>
|
||||
stream
|
||||
Gas3-b=]`-&-h'@TAj4XMk6`hV:j"r+Qu/5_JPH2[*jl@3?0-u[(9'GR:-/*/ft>A_=nj;i7d2;EpsDoJr<OBhVlHiq4E/El7+06*H?(h_eGnqiS:>Dgn0>N^CGqOd65m'$2XdN[8"CN<R^<p;O;.QTL>"4'o-s=`lHc!JpSi8$*d@]6l&@V%Q+V`W6/nPEL_rB?OF1iZbk.;Ju<];RLo@-9lO$dQ,9&`I`%EM@\dBr0Lf$$+R^&+/ncK?;0=7o:`];ceF"uKA7ETdrT"0YNT=QC"`>/@%I83@M@]K&@Nk~>endstream
|
||||
endobj
|
||||
xref
|
||||
0 10
|
||||
0000000000 65535 f
|
||||
0000000073 00000 n
|
||||
0000000104 00000 n
|
||||
0000000211 00000 n
|
||||
0000004431 00000 n
|
||||
0000009270 00000 n
|
||||
0000009574 00000 n
|
||||
0000009642 00000 n
|
||||
0000009938 00000 n
|
||||
0000009997 00000 n
|
||||
trailer
|
||||
<<
|
||||
/ID
|
||||
[<60f7c7338a7d1cfd54f86e6a06e41602><60f7c7338a7d1cfd54f86e6a06e41602>]
|
||||
% ReportLab generated PDF document -- digest (http://www.reportlab.com)
|
||||
|
||||
/Info 7 0 R
|
||||
/Root 6 0 R
|
||||
/Size 10
|
||||
>>
|
||||
startxref
|
||||
10387
|
||||
%%EOF
|
||||
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
@@ -0,0 +1,223 @@
|
||||
"""
|
||||
Unit tests for DocxConverterWithOCR.
|
||||
|
||||
For each DOCX test file: convert with a mock OCR service then compare the
|
||||
full output string against the expected snapshot.
|
||||
|
||||
OCR block format used by the converter:
|
||||
*[Image OCR]
|
||||
MOCK_OCR_TEXT_12345
|
||||
[End OCR]*
|
||||
"""
|
||||
|
||||
import sys
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
import pytest
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
|
||||
|
||||
from markitdown_ocr._ocr_service import OCRResult # noqa: E402
|
||||
from markitdown_ocr._docx_converter_with_ocr import ( # noqa: E402
|
||||
DocxConverterWithOCR,
|
||||
)
|
||||
from markitdown import StreamInfo # noqa: E402
|
||||
|
||||
TEST_DATA_DIR = Path(__file__).parent / "ocr_test_data"
|
||||
|
||||
_MOCK_TEXT = "MOCK_OCR_TEXT_12345"
|
||||
|
||||
|
||||
class MockOCRService:
|
||||
def extract_text( # noqa: ANN101
|
||||
self, image_stream: Any, **kwargs: Any
|
||||
) -> OCRResult:
|
||||
return OCRResult(text=_MOCK_TEXT, backend_used="mock")
|
||||
|
||||
|
||||
@pytest.fixture(scope="module")
|
||||
def svc() -> MockOCRService:
|
||||
return MockOCRService()
|
||||
|
||||
|
||||
def _convert(filename: str, ocr_service: MockOCRService) -> str:
|
||||
path = TEST_DATA_DIR / filename
|
||||
if not path.exists():
|
||||
pytest.skip(f"Test file not found: {path}")
|
||||
converter = DocxConverterWithOCR()
|
||||
with open(path, "rb") as f:
|
||||
return converter.convert(
|
||||
f, StreamInfo(extension=".docx"), ocr_service=ocr_service
|
||||
).text_content
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# docx_image_start.docx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_docx_image_start(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"Document with Image at Start\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"This is the main content after the header image.\n\n"
|
||||
"More text content here."
|
||||
)
|
||||
assert _convert("docx_image_start.docx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# docx_image_middle.docx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_docx_image_middle(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"# Introduction\n\n"
|
||||
"This is the introduction section.\n\n"
|
||||
"We will see an image below.\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"# Analysis\n\n"
|
||||
"This section comes after the image."
|
||||
)
|
||||
assert _convert("docx_image_middle.docx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# docx_image_end.docx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_docx_image_end(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"Report\n\n"
|
||||
"Main findings of the report.\n\n"
|
||||
"Details and analysis.\n\n"
|
||||
"Recommendations.\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("docx_image_end.docx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# docx_multiple_images.docx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_docx_multiple_images(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"Multi-Image Document\n\n"
|
||||
"First section\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"Second section with another image\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"Conclusion"
|
||||
)
|
||||
assert _convert("docx_multiple_images.docx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# docx_multipage.docx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_docx_multipage(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"# Page 1 - Mixed Content\n\n"
|
||||
"This is the first paragraph on page 1.\n\n"
|
||||
"BEFORE IMAGE: Important content appears here.\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"AFTER IMAGE: This content follows the image.\n\n"
|
||||
"More text on page 1.\n\n"
|
||||
"# Page 2 - Image at End\n\n"
|
||||
"Content on page 2.\n\n"
|
||||
"Multiple paragraphs of text.\n\n"
|
||||
"Building up to the image...\n\n"
|
||||
"Final paragraph before image.\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"# Page 3 - Image at Start\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"Content that follows the header image.\n\n"
|
||||
"AFTER IMAGE: This text is after the image."
|
||||
)
|
||||
assert _convert("docx_multipage.docx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# docx_complex_layout.docx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_docx_complex_layout(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"Complex Document\n\n"
|
||||
"| | |\n"
|
||||
"| --- | --- |\n"
|
||||
"| Feature | Status |\n"
|
||||
"| Authentication | Active |\n"
|
||||
"| Encryption | Enabled |\n\n"
|
||||
"Security notice:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("docx_complex_layout.docx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# _inject_placeholders — internal unit tests (no file I/O)
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_inject_placeholders_single_image() -> None:
|
||||
converter = DocxConverterWithOCR()
|
||||
html = "<p>Before</p><img src='x.png'/><p>After</p>"
|
||||
result_html, texts = converter._inject_placeholders(html, {"rId1": "TEXT"})
|
||||
assert "<img" not in result_html
|
||||
assert "MARKITDOWNOCRBLOCK0" in result_html
|
||||
assert texts == ["TEXT"]
|
||||
|
||||
|
||||
def test_inject_placeholders_two_images_sequential_tokens() -> None:
|
||||
converter = DocxConverterWithOCR()
|
||||
html = "<img src='a.png'/><p>Mid</p><img src='b.png'/>"
|
||||
result_html, texts = converter._inject_placeholders(
|
||||
html, {"rId1": "FIRST", "rId2": "SECOND"}
|
||||
)
|
||||
assert "MARKITDOWNOCRBLOCK0" in result_html
|
||||
assert "MARKITDOWNOCRBLOCK1" in result_html
|
||||
assert result_html.index("MARKITDOWNOCRBLOCK0") < result_html.index(
|
||||
"MARKITDOWNOCRBLOCK1"
|
||||
)
|
||||
assert len(texts) == 2
|
||||
|
||||
|
||||
def test_inject_placeholders_no_img_tag_appends_at_end() -> None:
|
||||
converter = DocxConverterWithOCR()
|
||||
html = "<p>No images</p>"
|
||||
result_html, texts = converter._inject_placeholders(html, {"rId1": "ORPHAN"})
|
||||
assert "MARKITDOWNOCRBLOCK0" in result_html
|
||||
assert texts == ["ORPHAN"]
|
||||
|
||||
|
||||
def test_inject_placeholders_empty_map_leaves_html_unchanged() -> None:
|
||||
converter = DocxConverterWithOCR()
|
||||
html = "<p>Content</p><img src='pic.jpg'/>"
|
||||
result_html, texts = converter._inject_placeholders(html, {})
|
||||
assert result_html == html
|
||||
assert texts == []
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# No OCR service — no OCR tags emitted
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_docx_no_ocr_service_no_tags() -> None:
|
||||
path = TEST_DATA_DIR / "docx_image_middle.docx"
|
||||
if not path.exists():
|
||||
pytest.skip(f"Test file not found: {path}")
|
||||
converter = DocxConverterWithOCR()
|
||||
with open(path, "rb") as f:
|
||||
md = converter.convert(f, StreamInfo(extension=".docx")).text_content
|
||||
assert "*[Image OCR]" not in md
|
||||
assert "[End OCR]*" not in md
|
||||
@@ -0,0 +1,234 @@
|
||||
"""
|
||||
Unit tests for PdfConverterWithOCR.
|
||||
|
||||
For each PDF test file: convert with a mock OCR service then compare the
|
||||
full output string against the expected snapshot.
|
||||
|
||||
OCR block format used by the converter:
|
||||
*[Image OCR]
|
||||
MOCK_OCR_TEXT_12345
|
||||
[End OCR]*
|
||||
"""
|
||||
|
||||
import io
|
||||
import sys
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
from unittest.mock import MagicMock, patch
|
||||
|
||||
import pytest
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
|
||||
|
||||
from markitdown_ocr._ocr_service import OCRResult # noqa: E402
|
||||
from markitdown_ocr._pdf_converter_with_ocr import ( # noqa: E402
|
||||
PdfConverterWithOCR,
|
||||
)
|
||||
from markitdown import StreamInfo # noqa: E402
|
||||
|
||||
TEST_DATA_DIR = Path(__file__).parent / "ocr_test_data"
|
||||
|
||||
_MOCK_TEXT = "MOCK_OCR_TEXT_12345"
|
||||
_OCR_BLOCK = f"*[Image OCR]\n{_MOCK_TEXT}\n[End OCR]*"
|
||||
_PAGE_1_SCANNED = f"## Page 1\n\n\n\n\n{_OCR_BLOCK}"
|
||||
|
||||
|
||||
class MockOCRService:
|
||||
def extract_text(
|
||||
self, # noqa: ANN101
|
||||
image_stream: Any,
|
||||
**kwargs: Any,
|
||||
) -> OCRResult:
|
||||
return OCRResult(text=_MOCK_TEXT, backend_used="mock")
|
||||
|
||||
|
||||
@pytest.fixture(scope="module")
|
||||
def svc() -> MockOCRService:
|
||||
return MockOCRService()
|
||||
|
||||
|
||||
def _convert(filename: str, ocr_service: MockOCRService) -> str:
|
||||
path = TEST_DATA_DIR / filename
|
||||
if not path.exists():
|
||||
pytest.skip(f"Test file not found: {path}")
|
||||
converter = PdfConverterWithOCR()
|
||||
with open(path, "rb") as f:
|
||||
return converter.convert(
|
||||
f, StreamInfo(extension=".pdf"), ocr_service=ocr_service
|
||||
).text_content
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pdf_image_start.pdf
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pdf_image_start(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Page 1\n\n\n\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n\n"
|
||||
"This is text BEFORE the image.\n\n"
|
||||
"The image should appear above this text.\n\n"
|
||||
"This is more content after the image."
|
||||
)
|
||||
assert _convert("pdf_image_start.pdf", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pdf_image_middle.pdf
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pdf_image_middle(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Page 1\n\n\n"
|
||||
"Section 1: Introduction\n\n"
|
||||
"This document contains an image in the middle.\n\n"
|
||||
"Here is some introductory text.\n\n\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n\n"
|
||||
"Section 2: Details\n\n"
|
||||
"This text appears AFTER the image."
|
||||
)
|
||||
assert _convert("pdf_image_middle.pdf", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pdf_image_end.pdf
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pdf_image_end(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Page 1\n\n\n"
|
||||
"Main Content\n\n"
|
||||
"This is the main text content.\n\n"
|
||||
"The image will appear at the end.\n\n"
|
||||
"Keep reading...\n\n\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("pdf_image_end.pdf", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pdf_multiple_images.pdf
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pdf_multiple_images(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Page 1\n\n\n"
|
||||
"Document with Multiple Images\n\n\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n\n"
|
||||
"Text between first and second image.\n\n\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n\n"
|
||||
"Final text after all images."
|
||||
)
|
||||
assert _convert("pdf_multiple_images.pdf", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pdf_complex_layout.pdf
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pdf_complex_layout(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Page 1\n\n\n"
|
||||
"Complex Layout Document\n\n"
|
||||
"Table:\n\n"
|
||||
"ItemQuantity\n\n\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n\n"
|
||||
"Widget A5"
|
||||
)
|
||||
assert _convert("pdf_complex_layout.pdf", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pdf_multipage.pdf — pdfplumber/pdfminer fail (EOF); PyMuPDF fallback used
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pdf_multipage(svc: MockOCRService) -> None:
|
||||
# pdfplumber cannot open this file (Unexpected EOF), so _ocr_full_pages
|
||||
# falls back to PyMuPDF for page rendering. Each page becomes one OCR block.
|
||||
expected = (
|
||||
f"## Page 1\n\n\n{_OCR_BLOCK}\n\n\n"
|
||||
f"## Page 2\n\n\n{_OCR_BLOCK}\n\n\n"
|
||||
f"## Page 3\n\n\n{_OCR_BLOCK}"
|
||||
)
|
||||
assert _convert("pdf_multipage.pdf", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pdf_scanned_*.pdf — raster-only pages → full-page OCR
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pdf_scanned_invoice(svc: MockOCRService) -> None:
|
||||
assert _convert("pdf_scanned_invoice.pdf", svc) == _PAGE_1_SCANNED
|
||||
|
||||
|
||||
def test_pdf_scanned_meeting_minutes(svc: MockOCRService) -> None:
|
||||
assert _convert("pdf_scanned_meeting_minutes.pdf", svc) == _PAGE_1_SCANNED
|
||||
|
||||
|
||||
def test_pdf_scanned_minimal(svc: MockOCRService) -> None:
|
||||
assert _convert("pdf_scanned_minimal.pdf", svc) == _PAGE_1_SCANNED
|
||||
|
||||
|
||||
def test_pdf_scanned_sales_report(svc: MockOCRService) -> None:
|
||||
assert _convert("pdf_scanned_sales_report.pdf", svc) == _PAGE_1_SCANNED
|
||||
|
||||
|
||||
def test_pdf_scanned_report(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
f"{_PAGE_1_SCANNED}\n\n\n\n"
|
||||
f"## Page 2\n\n\n\n\n{_OCR_BLOCK}\n\n\n\n"
|
||||
f"## Page 3\n\n\n\n\n{_OCR_BLOCK}"
|
||||
)
|
||||
assert _convert("pdf_scanned_report.pdf", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Scanned PDF fallback path (pdfplumber finds no text → full-page OCR)
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pdf_scanned_fallback_format(svc: MockOCRService) -> None:
|
||||
"""_ocr_full_pages emits *[Image OCR]...[End OCR]* for each page."""
|
||||
path = TEST_DATA_DIR / "pdf_image_start.pdf"
|
||||
if not path.exists():
|
||||
pytest.skip(f"Test file not found: {path}")
|
||||
|
||||
converter = PdfConverterWithOCR()
|
||||
with patch("pdfplumber.open") as mock_plumber:
|
||||
mock_pdf = MagicMock()
|
||||
mock_page = MagicMock()
|
||||
mock_page.page_number = 1
|
||||
mock_pdf.pages = [mock_page]
|
||||
mock_pdf.__enter__.return_value = mock_pdf
|
||||
mock_plumber.return_value = mock_pdf
|
||||
|
||||
with open(path, "rb") as f:
|
||||
md = converter._ocr_full_pages(io.BytesIO(f.read()), svc)
|
||||
|
||||
expected = "## Page 1\n\n\n" "*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
assert (
|
||||
md == expected
|
||||
), f"_ocr_full_pages must produce:\n{expected!r}\nActual:\n{md!r}"
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# No OCR service — no OCR tags emitted
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pdf_no_ocr_service_no_tags() -> None:
|
||||
path = TEST_DATA_DIR / "pdf_image_middle.pdf"
|
||||
if not path.exists():
|
||||
pytest.skip(f"Test file not found: {path}")
|
||||
converter = PdfConverterWithOCR()
|
||||
with open(path, "rb") as f:
|
||||
md = converter.convert(f, StreamInfo(extension=".pdf")).text_content
|
||||
assert "*[Image OCR]" not in md
|
||||
assert "[End OCR]*" not in md
|
||||
@@ -0,0 +1,148 @@
|
||||
"""
|
||||
Unit tests for PptxConverterWithOCR.
|
||||
|
||||
For each PPTX test file: convert with a mock OCR service then compare the
|
||||
full output string against the expected snapshot.
|
||||
|
||||
OCR block format used by the converter:
|
||||
*[Image OCR]
|
||||
MOCK_OCR_TEXT_12345
|
||||
[End OCR]*
|
||||
|
||||
Note: PPTX slide text uses literal backslash-n (\\n) sequences from the
|
||||
underlying PPTX converter template; OCR blocks use real newlines.
|
||||
"""
|
||||
|
||||
import sys
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
import pytest
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
|
||||
|
||||
from markitdown_ocr._ocr_service import OCRResult # noqa: E402
|
||||
from markitdown_ocr._pptx_converter_with_ocr import ( # noqa: E402
|
||||
PptxConverterWithOCR,
|
||||
)
|
||||
from markitdown import StreamInfo # noqa: E402
|
||||
|
||||
TEST_DATA_DIR = Path(__file__).parent / "ocr_test_data"
|
||||
|
||||
_MOCK_TEXT = "MOCK_OCR_TEXT_12345"
|
||||
_OCR_BLOCK = f"*[Image OCR]\n{_MOCK_TEXT}\n[End OCR]*"
|
||||
|
||||
|
||||
class MockOCRService:
|
||||
def extract_text(
|
||||
self, # noqa: ANN101
|
||||
image_stream: Any,
|
||||
**kwargs: Any,
|
||||
) -> OCRResult:
|
||||
return OCRResult(text=_MOCK_TEXT, backend_used="mock")
|
||||
|
||||
|
||||
@pytest.fixture(scope="module")
|
||||
def svc() -> MockOCRService:
|
||||
return MockOCRService()
|
||||
|
||||
|
||||
def _convert(filename: str, ocr_service: MockOCRService) -> str:
|
||||
path = TEST_DATA_DIR / filename
|
||||
if not path.exists():
|
||||
pytest.skip(f"Test file not found: {path}")
|
||||
converter = PptxConverterWithOCR()
|
||||
with open(path, "rb") as f:
|
||||
return converter.convert(
|
||||
f, StreamInfo(extension=".pptx"), ocr_service=ocr_service
|
||||
).text_content
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pptx_image_start.pptx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pptx_image_start(svc: MockOCRService) -> None:
|
||||
# Slide 1: title "Welcome" followed by an image
|
||||
expected = (
|
||||
"\\n\\n<!-- Slide number: 1 -->\\n# Welcome\\n\\n"
|
||||
"\n*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("pptx_image_start.pptx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pptx_image_middle.pptx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pptx_image_middle(svc: MockOCRService) -> None:
|
||||
# Slide 1: Introduction | Slide 2: Architecture + image | Slide 3: Conclusion # noqa: E501
|
||||
expected = (
|
||||
"\\n\\n<!-- Slide number: 1 -->\\n# Introduction"
|
||||
"\\n\\n\\n\\n<!-- Slide number: 2 -->\\n# Architecture\\n\\n"
|
||||
"\n*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
"\\n\\n<!-- Slide number: 3 -->\\n# Conclusion\\n\\n"
|
||||
)
|
||||
assert _convert("pptx_image_middle.pptx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pptx_image_end.pptx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pptx_image_end(svc: MockOCRService) -> None:
|
||||
# Slide 1: Presentation | Slide 2: Thank You + image
|
||||
expected = (
|
||||
"\\n\\n<!-- Slide number: 1 -->\\n# Presentation"
|
||||
"\\n\\n\\n\\n<!-- Slide number: 2 -->\\n# Thank You\\n\\n"
|
||||
"\n*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("pptx_image_end.pptx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pptx_multiple_images.pptx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pptx_multiple_images(svc: MockOCRService) -> None:
|
||||
# Slide 1: two images, no title text
|
||||
expected = (
|
||||
"\\n\\n<!-- Slide number: 1 -->\\n# \\n"
|
||||
"\n*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
"\n\n*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("pptx_multiple_images.pptx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# pptx_complex_layout.pptx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pptx_complex_layout(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"\\n\\n<!-- Slide number: 1 -->\\n# Product Comparison"
|
||||
"\\n\\nOur products lead the market\\n"
|
||||
"\n*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("pptx_complex_layout.pptx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# No OCR service — no OCR tags emitted
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_pptx_no_ocr_service_no_tags() -> None:
|
||||
path = TEST_DATA_DIR / "pptx_image_middle.pptx"
|
||||
if not path.exists():
|
||||
pytest.skip(f"Test file not found: {path}")
|
||||
converter = PptxConverterWithOCR()
|
||||
with open(path, "rb") as f:
|
||||
md = converter.convert(f, StreamInfo(extension=".pptx")).text_content
|
||||
assert "*[Image OCR]" not in md
|
||||
assert "[End OCR]*" not in md
|
||||
@@ -0,0 +1,249 @@
|
||||
"""
|
||||
Unit tests for XlsxConverterWithOCR.
|
||||
|
||||
For each XLSX test file: convert with a mock OCR service then compare the
|
||||
full output string against the expected snapshot.
|
||||
|
||||
OCR block format used by the converter:
|
||||
*[Image OCR]
|
||||
MOCK_OCR_TEXT_12345
|
||||
[End OCR]*
|
||||
|
||||
Images are grouped at the end of each sheet under:
|
||||
### Images in this sheet:
|
||||
"""
|
||||
|
||||
import sys
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
import pytest
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
|
||||
|
||||
from markitdown_ocr._ocr_service import OCRResult # noqa: E402
|
||||
from markitdown_ocr._xlsx_converter_with_ocr import ( # noqa: E402
|
||||
XlsxConverterWithOCR,
|
||||
)
|
||||
from markitdown import StreamInfo # noqa: E402
|
||||
|
||||
TEST_DATA_DIR = Path(__file__).parent / "ocr_test_data"
|
||||
|
||||
_MOCK_TEXT = "MOCK_OCR_TEXT_12345"
|
||||
_OCR_BLOCK = f"*[Image OCR]\n{_MOCK_TEXT}\n[End OCR]*"
|
||||
_IMG_SECTION = "### Images in this sheet:"
|
||||
|
||||
|
||||
class MockOCRService:
|
||||
def extract_text(
|
||||
self, # noqa: ANN101
|
||||
image_stream: Any,
|
||||
**kwargs: Any,
|
||||
) -> OCRResult:
|
||||
return OCRResult(text=_MOCK_TEXT, backend_used="mock")
|
||||
|
||||
|
||||
@pytest.fixture(scope="module")
|
||||
def svc() -> MockOCRService:
|
||||
return MockOCRService()
|
||||
|
||||
|
||||
def _convert(filename: str, ocr_service: MockOCRService) -> str:
|
||||
path = TEST_DATA_DIR / filename
|
||||
if not path.exists():
|
||||
pytest.skip(f"Test file not found: {path}")
|
||||
converter = XlsxConverterWithOCR()
|
||||
with open(path, "rb") as f:
|
||||
return converter.convert(
|
||||
f, StreamInfo(extension=".xlsx"), ocr_service=ocr_service
|
||||
).text_content
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# xlsx_image_start.xlsx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_xlsx_image_start(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Sales Q1\n\n"
|
||||
"| Product | Sales |\n"
|
||||
"| --- | --- |\n"
|
||||
"| Widget A | 100 |\n"
|
||||
"| Widget B | 150 |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"## Forecast Q2\n\n"
|
||||
"| Projected Sales | Unnamed: 1 |\n"
|
||||
"| --- | --- |\n"
|
||||
"| Widget A | 120 |\n"
|
||||
"| Widget B | 180 |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("xlsx_image_start.xlsx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# xlsx_image_middle.xlsx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_xlsx_image_middle(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Revenue\n\n"
|
||||
"| Q1 Report | Unnamed: 1 |\n"
|
||||
"| --- | --- |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| Revenue | $50,000 |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| Profit Margin | 40% |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"## Expenses\n\n"
|
||||
"| Expense Breakdown | Unnamed: 1 |\n"
|
||||
"| --- | --- |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| Expenses | $30,000 |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| Savings | $5,000 |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("xlsx_image_middle.xlsx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# xlsx_image_end.xlsx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_xlsx_image_end(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Sheet\n\n"
|
||||
"| Financial Summary | Unnamed: 1 |\n"
|
||||
"| --- | --- |\n"
|
||||
"| Total Revenue | $500,000 |\n"
|
||||
"| Total Expenses | $300,000 |\n"
|
||||
"| Net Profit | $200,000 |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| Signature: | NaN |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"## Budget\n\n"
|
||||
"| Budget Allocation | Unnamed: 1 |\n"
|
||||
"| --- | --- |\n"
|
||||
"| Marketing | $100,000 |\n"
|
||||
"| R&D | $150,000 |\n"
|
||||
"| Operations | $50,000 |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| Approved: | NaN |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("xlsx_image_end.xlsx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# xlsx_multiple_images.xlsx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_xlsx_multiple_images(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Overview\n\n"
|
||||
"| Dashboard |\n"
|
||||
"| --- |\n"
|
||||
"| Status: Active |\n"
|
||||
"| NaN |\n"
|
||||
"| NaN |\n"
|
||||
"| NaN |\n"
|
||||
"| NaN |\n"
|
||||
"| Performance Summary |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"## Details\n\n"
|
||||
"| Detailed Metrics |\n"
|
||||
"| --- |\n"
|
||||
"| System Health |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"## Summary\n\n"
|
||||
"| Quarter Summary |\n"
|
||||
"| --- |\n"
|
||||
"| Overall Performance |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("xlsx_multiple_images.xlsx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# xlsx_complex_layout.xlsx
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_xlsx_complex_layout(svc: MockOCRService) -> None:
|
||||
expected = (
|
||||
"## Complex Report\n\n"
|
||||
"| Annual Report 2024 | Unnamed: 1 |\n"
|
||||
"| --- | --- |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| Month | Sales |\n"
|
||||
"| Jan | 1000 |\n"
|
||||
"| Feb | 1200 |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| Total | 2200 |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"## Customers\n\n"
|
||||
"| Customer Metrics | Unnamed: 1 |\n"
|
||||
"| --- | --- |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| New Customers | 250 |\n"
|
||||
"| Retention Rate | 92% |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*\n\n"
|
||||
"## Regions\n\n"
|
||||
"| Regional Breakdown | Unnamed: 1 |\n"
|
||||
"| --- | --- |\n"
|
||||
"| NaN | NaN |\n"
|
||||
"| Region | Revenue |\n"
|
||||
"| North | $800K |\n"
|
||||
"| South | $600K |\n\n"
|
||||
"### Images in this sheet:\n\n"
|
||||
"*[Image OCR]\nMOCK_OCR_TEXT_12345\n[End OCR]*"
|
||||
)
|
||||
assert _convert("xlsx_complex_layout.xlsx", svc) == expected
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# No OCR service — no OCR tags emitted
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
def test_xlsx_no_ocr_service_no_tags() -> None:
|
||||
path = TEST_DATA_DIR / "xlsx_image_middle.xlsx"
|
||||
if not path.exists():
|
||||
pytest.skip(f"Test file not found: {path}")
|
||||
converter = XlsxConverterWithOCR()
|
||||
with open(path, "rb") as f:
|
||||
md = converter.convert(f, StreamInfo(extension=".xlsx")).text_content
|
||||
assert "*[Image OCR]" not in md
|
||||
assert "[End OCR]*" not in md
|
||||
@@ -1,4 +1,4 @@
|
||||
# SPDX-FileCopyrightText: 2024-present Adam Fourney <adamfo@microsoft.com>
|
||||
#
|
||||
# SPDX-License-Identifier: MIT
|
||||
__version__ = "0.1.5"
|
||||
__version__ = "0.1.6b1"
|
||||
|
||||
Reference in New Issue
Block a user