Page Flow Model for OCR Text Area Coordinate Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional OCR systems fail to accurately process documents with variable length text areas, leading to page misalignment and errors when text overflows from one page to another, as they are not trained to handle shifts in text area coordinates.
Innovation Solution
A computer-implemented method that generates text area data based on font type and size, trains a page flow model to predict potential overflows or underflows, and adjusts coordinate locations using a scanning script to maintain accurate OCR processing across multiple pages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional OCR systems process documents with variable length text areas, then basic OCR functionality is provided, but page misalignment and errors occur when text overflows from one page to another
Solution Approach 1:
The system performs preliminary actions by training an OCR model in advance using synthetic documents generated with various text area configurations, including overflow scenarios. This pre-training enables the model to anticipate and correctly handle text overflow situations during actual OCR processing without requiring real-time adjustments
Solution Approach 2:
The system dynamically adjusts text area coordinates based on detected text content and overflow patterns. The OCR model learns to adapt coordinate positions dynamically according to the actual text length and distribution, allowing accurate text recognition even when text overflows across page boundaries
2Adaptability or versatility
If text areas are fixed in coordinate locations, then OCR processing is straightforward, but text areas cannot adapt to variable length content causing page shifts
Solution Approach 1:
The system pre-generates synthetic documents with variable length text areas and trains the OCR model on these diverse examples. This preliminary training establishes a foundation of coordinate patterns that the model can recognize and adapt to, enabling flexible text area handling while maintaining coordinate precision
Solution Approach 2:
The system changes coordinate parameters dynamically based on text content characteristics. The OCR model learns to adjust text area coordinates as a function of text length, font size, and content distribution, allowing the same document template to accommodate variable content while maintaining accurate text recognition
Data Source
AI summary
Provided are techniques for OCR induction for multi-page changes. A plurality of documents of a document type are processed to generate text area data for a text area in one or more documents of the plurality of documents, where the text area data includes coordinate locations of a zone for the text area based on expansion and direction of shift of the text area. A page flow model is trained using the plurality of documents and the text area data. In response to receiving a new document comprising the text area, a scanning script is received from the page flow model, where the page flow model identifies a new zone for the text area in the new document and determines how to adjust another zone for an element in the new document. The scanning script is used to scan the new document to generate digital text.