Amazon Textract Textractor

Textractor is a Python package designed to simplify working with four key Amazon Textract APIs: DocumentTextDetection, StartDocumentTextDetection, AnalyzeDocument, and StartDocumentAnalysis.

It provides utilities to:

• Call Textract services with reduced parameter overhead.
• Convert JSON responses from Textract APIs into programmable objects.
• Visualize entities directly on document images.
• Export document data into compatible formats like Excel, CSV, and TXT.

Key advanced features include:

• Semantic Document Search.
• Key-value querying using specific keys.
• Table access using numpy indexing.
• Detection of duplicated document entities.
• Functionality available at both the Document and Page levels.

Textractor models the document as a hierarchy of entities. You can traverse from the top-level Document down to individual characters or words:

• Page: Represents a single page within a multi-page document.
• Line: Represents a line of text.
• Word: Represents an individual word.

All these entities inherit from or are related to DocumentEntity and typically include spatial information via BoundingBox.

For tabular data extraction, Textractor provides specific classes to handle the structure of tables:

• Table: Represents a complete table entity.
• TableCell: Represents an individual cell within a table.
• TableTitle: Represents the title/header text associated with a table.
• TableFooter: Represents footer information within a table.

Textractor provides programmatic classes to access information produced by Amazon Textract analysis APIs. The package maps Textract's recognized document entities into structured Python objects.

Textract recognizes 6 primary document entities which are represented in Textractor as:

• WORD
• LINE
• KEY_VALUE_SET
• SELECTION_ELEMENT
• TABLE
• CELL

These entities are contained within Document objects, allowing you to navigate the document structure (e.g., accessing pages, lines, or tables) using specialized entity classes.

Textractor includes high-level entities for specific document types processed by Amazon Textract:

Expense Documents:

• ExpenseDocument: The top-level container for expense analysis.
• Expense: Represents individual expense line items or fields.

Identity Documents:

• IdentityDocument: The top-level container for identity document analysis.
• IdentityField: Represents specific fields within an identity document (e.g., Name, ID Number).

The Document class is the primary container for all entities extracted from a document. It holds the hierarchical structure of the analysis results.

LazyDocument is an alternative class (likely used for deferred loading or memory efficiency) that provides access to the same underlying document entities.

Instead of using hardcoded x,y coordinates, amazon-textract-geofinder uses the concept of an area.

An area is defined by a bounding box (x_min, y_min, x_max, y_max). You can dynamically create these areas by finding specific words or phrases in the document. This allows you to associate key/value pairs with specific sections (e.g., associating all fields found between the 'Patient Information' header and the 'Emergency Contact' header as 'Patient' fields).

Textractor provides specialized entities for form-based extraction:

• KeyValue: Represents a key-value pair (e.g., 'Name: John Doe').
• Value: Represents the value part of a key-value pair.
• SelectionElement: Represents checkboxes or radio buttons (selection elements) found in documents.

Install the helper package via pip to access the amazon-textract command line tool. Ensure your environment is configured with AWS credentials (via config files, environment variables, or IAM roles) before running commands.

python -m pip install amazon-textract-helper

Use the --pretty-print flag to format the JSON output into human-readable text for WORDS, LINES, FORMS, or TABLES. For tables, you can specify a format using --pretty-print-table-format (e.g., csv, grid, html).

# Pretty print tables
amazon-textract --example --features TABLES --pretty-print TABLES

# Pretty print both forms and tables
amazon-textract --example --features FORMS TABLES --pretty-print FORMS TABLES