Slug: splitter-rotator-classifier Version: 1.0.0 Infer: Yes

Overview

Splitter Rotator Classifier Model

This model provides advanced document preprocessing capabilities to prepare documents for extraction or further analysis. It can split multi-document scans, correct page orientation, filter blank or duplicate pages, and classify documents by type.

How It Works

The model performs a comprehensive set of document preprocessing operations:

Document Splitting: Detects when multiple documents appear on a single scanned page and separates them
Rotation Correction: Automatically rotates pages to the correct orientation
Page Filtering: Removes blank pages and duplicates to clean up the document set
Page Reordering: Intelligently groups and orders pages that belong to multi-page documents
Document Classification: Identifies document types based on content and layout

Options Configuration

Option	Description
document_types	List of document types to classify against (can include hints in parentheses)
split_multiple_documents_in_one_page	When enabled, detects and splits pages containing multiple distinct documents
correct_rotation	Analyzes and corrects page orientation for proper reading
exclude_blank_pages	Filters out completely blank or mostly white pages
exclude_duplicate_pages	Removes duplicate pages based on visual similarity
reorder_pages	Intelligently groups pages that belong to the same multi-page document
multiple_document_in_one_page_identifier_model	LLM model used to verify if a page contains multiple documents
document_classifier_model	LLM model used to determine document types
multiple_page_document_identifier_model	LLM model used to identify pages belonging to the same document
target_store	Document store to receive the processed documents

Process Flow

Document Splitting Process

When splitting multiple documents in a single page:

Image Analysis: The model scans for whitespace gaps or content boundaries
Split Point Detection: It identifies potential horizontal (stacked) or vertical (side-by-side) split points
LLM Validation: An AI model verifies that the detected split represents two distinct documents
Document Separation: The page is split into separate documents at the validated boundary

Classification Process

Document classification uses AI to identify document types:

The model analyzes page content, layout, and text
It compares against your defined document types (with optional hints)
For multi-page documents, it considers content across all pages
Each document is assigned the most likely document type

Example Usage

This model is particularly useful for:

Processing batch scans that contain multiple different documents
Handling document collections with mixed orientations and types
Cleaning up scans by removing blank separator pages
Organizing multi-page documents that were scanned out of order
Automatically categorizing documents for downstream processing

Configuration Example

To process mixed-orientation invoice scans:

document_types: "- Invoice (hints: total, amount due, bill to)\n- Receipt (hints: paid, payment, thank you)\n- Packing Slip (hints: shipped items, quantity)"
split_multiple_documents_in_one_page: true
correct_rotation: true
exclude_blank_pages: true
exclude_duplicate_pages: true
reorder_pages: true
document_classifier_model: "gemini-2.0-flash-001"

Inference Options

The following options can be configured when using this model for inference:

Name	Label	Type	Description	Default	Required
`document_types`	Document Types	string	The possible document types to use for classification. You can include hints in parentheses.	- Document Type 1 (hints:)	No
`split_multiple_documents_in_one_page`	Split multiple documents in one page?	boolean	Whether to split multiple documents in one page	True	No
`correct_rotation`	Correct rotation of documents?	boolean	Whether to correct the rotation of the document	True	No
`exclude_blank_pages`	Exclude blank pages?	boolean	Whether to exclude blank pages	True	No
`exclude_duplicate_pages`	Exclude duplicate pages?	boolean	Whether to exclude duplicate pages	True	No
`reorder_pages`	Reorder pages?	boolean	Whether to reorder pages based on the multi-page results	True	No
`multiple_document_in_one_page_identifier_model`	Multiple documents in one page identifier LLM model	string	The model to use to identify pages with multiple documents scanned into them	gemini-2.0-flash-001	No
`document_classifier_model`	Document classifier LLM model	string	The model to use to classify documents	gemini-2.0-flash-001	No
`multiple_page_document_identifier_model`	Multiple page document identifier LLM model	string	The model to use to identify documents that are scanned across multiple pages	gemini-2.0-flash-001	No
`target_store`	Target Store	documentStore	The store that should receive the processed document	-	No

Model Details

Provider: Kodexa

Introduction

​Overview

​Splitter Rotator Classifier Model

​How It Works

​Options Configuration

​Process Flow

​Document Splitting Process

​Classification Process

​Example Usage

​Configuration Example

​Inference Options

​Model Details