Azure Document Intelligence Loader¶

The Azure Document Intelligence loader (formerly Form Recognizer) uses Azure's Document Intelligence service to extract text, tables, and structured information from documents.

Supported Formats¶

Supports PDF, JPEG/JPG, PNG, BMP, TIFF, HEIF, DOCX, XLSX, PPTX and HTML.

Usage¶

Basic Usage¶

from extract_thinker import DocumentLoaderAzureForm

# Initialize with Azure credentials
loader = DocumentLoaderAzureForm(
    subscription_key="your_subscription_key",
    endpoint="your_endpoint",
    model_id="prebuilt-document"  # Use prebuilt document model
)

# Load document
pages = loader.load("path/to/your/document.pdf")

# Process extracted content
for page in pages:
    # Access text content
    text = page["content"]
    # Access tables if available
    tables = page.get("tables", [])
    # Access form fields if available
    forms = page.get("forms", {})

Configuration-based Usage¶

from extract_thinker import DocumentLoaderAzureForm, AzureConfig

# Create configuration
config = AzureConfig(
    subscription_key="your_subscription_key",
    endpoint="your_endpoint",
    model_id="prebuilt-layout",     # Use layout model for enhanced layout analysis
    cache_ttl=600,                  # Cache results for 10 minutes
    features=["ocrHighResolution", "barcodes"]  # Enable advanced features
)

# Initialize loader with configuration
loader = DocumentLoaderAzureForm(config)

# Load and process document
pages = loader.load("path/to/your/document.pdf")

Advanced Features Usage¶

from extract_thinker import DocumentLoaderAzureForm, AzureConfig

# Configuration with multiple advanced features
config = AzureConfig(
    subscription_key="your_subscription_key",
    endpoint="your_endpoint",
    model_id="prebuilt-layout",
    features=[
        "ocrHighResolution",    # High resolution OCR for small text
        "formulas",             # Extract mathematical formulas in LaTeX
        "styleFont",            # Extract font properties
        "barcodes",             # Extract barcodes and QR codes
        "languages",            # Detect document languages
        "keyValuePairs"         # Extract key-value pairs from forms
    ]
)

loader = DocumentLoaderAzureForm(config)
pages = loader.load("document_with_advanced_content.pdf")

for page in pages:
    # Standard content
    print(f"Text content: {page['content']}")
    print(f"Tables: {page['tables']}")
    print(f"Forms: {page['forms']}")

    # Advanced features (if detected in document)
    if 'formulas' in page:
        print(f"Mathematical formulas: {page['formulas']}")

    if 'fonts' in page:
        print(f"Font information: {page['fonts']}")

    if 'barcodes' in page:
        print(f"Barcodes found: {page['barcodes']}")

    if 'languages' in page:
        print(f"Detected languages: {page['languages']}")

Specialized Models Usage¶

# Use specialized invoice model
config = AzureConfig(
    subscription_key="your_subscription_key",
    endpoint="your_endpoint",
    model_id="prebuilt-invoice"
)

loader = DocumentLoaderAzureForm(config)
pages = loader.load("invoice.pdf")

# Access extracted invoice fields
for page in pages:
    forms = page["forms"]
    vendor_name = forms.get("VendorName", "")
    invoice_total = forms.get("InvoiceTotal", "")
    print(f"Vendor: {vendor_name}, Total: {invoice_total}")

Configuration Options¶

The AzureConfig class supports the following options:

Option	Type	Default	Description
`subscription_key`	str	Required	Azure subscription key
`endpoint`	str	Required	Azure endpoint URL
`content`	Any	None	Initial content to process
`cache_ttl`	int	300	Cache time-to-live in seconds
`model_id`	str	"prebuilt-layout"	Model ID to use
`max_retries`	int	3	Maximum retries for failed requests
`features`	List[str]	None	Advanced features to enable

Available Models¶

General Purpose Models¶

Model ID	Description	Best For
`prebuilt-read`	OCR/Read model	Text extraction from printed and handwritten documents
`prebuilt-layout`	Layout analysis	Documents with tables, selection marks, and complex layouts
`prebuilt-document`	General document	Key-value pairs, tables, and general document structure

Specialized Models¶

Model ID	Description
`prebuilt-invoice`	Invoice processing
`prebuilt-receipt`	Receipt processing
`prebuilt-idDocument`	Identity documents
`prebuilt-businessCard`	Business cards
`prebuilt-tax.us.w2`	US W2 tax forms
`prebuilt-tax.us.1040`	US 1040 tax forms
`prebuilt-contract`	Contracts
`prebuilt-healthInsurance`	US health insurance cards
`prebuilt-bankStatement`	Bank statements
`prebuilt-payStub`	Pay stubs

Advanced Features¶

The loader supports advanced extraction features that can be enabled via the features parameter:

Feature	Description	Output Field
`ocrHighResolution`	High resolution OCR for better small text recognition	Enhanced text in `content`
`formulas`	Extract mathematical formulas in LaTeX format	`formulas` array
`styleFont`	Extract font properties (family, style, weight, color)	`fonts` array
`barcodes`	Extract barcodes and QR codes	`barcodes` array
`languages`	Detect document languages	`languages` array
`keyValuePairs`	Extract key-value pairs from forms	Enhanced `forms` dict
`queryFields`	Enable custom field extraction	Enhanced extraction
`searchablePDF`	Convert scanned PDFs to searchable format	Enhanced OCR

Features¶

Text extraction with layout preservation
Table detection and extraction
Form field recognition with specialized models
Advanced OCR with high resolution support
Mathematical formula extraction (LaTeX format)
Font property extraction
Barcode and QR code detection
Multi-language document support
Caching support with configurable TTL
Vision mode support for image formats
Retry logic for robust processing

Notes¶

Azure subscription key and endpoint are required
Advanced features may increase processing time and costs
Specialized models are optimized for specific document types
Rate limits and quotas apply based on your Azure subscription
Vision mode is supported for image formats
High resolution OCR is recommended for documents with small text
Formula extraction works best with clear mathematical notation

Select an API version¶

Set api_version to use a service version supported by the installed azure-ai-formrecognizer SDK. Omitting it preserves the SDK's default:

from extract_thinker import AzureConfig, DocumentLoaderAzureForm

config = AzureConfig(
    subscription_key="your_subscription_key",
    endpoint="https://your-resource.cognitiveservices.azure.com/",
    api_version="2022-08-31",
    model_id="prebuilt-layout",
)
loader = DocumentLoaderAzureForm(config)

The same option is accepted by the loader constructor and from_credentials. Models and advanced features must be compatible with the selected version. This loader uses azure-ai-formrecognizer; newer service APIs that require azure-ai-documentintelligence are not enabled simply by changing this string. See the Azure SDK version guidance.

Tables and empty cells¶

Each page's tables value is a list of tables; each table is a list of rows; each row is a list of cell strings. Missing cells retain an empty string in their original column. Multiple tables on a page are preserved separately. Merged-cell text is placed at its starting row/column; covered positions stay empty. A table spanning pages is associated with each page Azure reports.

config = AzureConfig(
    subscription_key="your_subscription_key",
    endpoint="https://your-resource.cognitiveservices.azure.com/",
    content_mode="tables",
)
loader = DocumentLoaderAzureForm(config)
pages = loader.load("invoice.pdf")

With content_mode="tables", content contains a JSON representation of the page's tables instead of surrounding OCR prose. The tables field remains available. Other metadata and explicitly enabled vision images remain available; use a text-only extraction if you do not want the original page image sent. The default "all" mode retains non-table text and removes exact duplicate cell lines from that text.

Migration: earlier versions flattened or overwrote tables on a page. Code that treated page["tables"] as rows should now iterate over tables first.