Document Bounding Boxes for OCR Token Mismatch Validation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional bounding box generation techniques struggle with accurately determining the location of extracted information that does not exactly match the document content, especially when the information includes multiple tokens or non-unique tokens, making it challenging to efficiently review and validate the accuracy of extracted data.

Innovation Solution

The proposed method generates bounding boxes by identifying OCR tokens in a document that are both typographically close and geometrically compact to the extracted information, using a trie data structure to enhance efficiency and accuracy in locating the correct positions for bounding box generation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional bounding box generation techniques are used, then the process is simple, but the accuracy of locating extracted information deteriorates when the information does not exactly match document content

Engineering Contradiction:

Improveaccuracy of bounding box locationVSAvoidcomplexity of bounding box generation process

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary process that compares extracted information with OCR token data to identify matching bounding boxes. This intermediary matching mechanism resolves the contradiction by enabling accurate location of extracted information without requiring exact string matches, thereby improving measurement precision while managing complexity through a structured comparison approach.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent segments the extracted information into individual tokens and compares each token against OCR token data separately. This segmentation allows the system to handle partial matches and non-unique tokens effectively, improving the accuracy of bounding box generation for complex cases without overwhelming the system with monolithic comparison operations.

Inventive Principle:

Principle #1Segmentation

2Adaptability or versatility

If exact matching is used to locate extracted information, then the process is straightforward, but the ability to handle non-unique tokens and partial matches deteriorates

Engineering Contradiction:

Improveability to handle non-unique tokens and partial matchesVSAvoidtime to review and validate extracted data

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent employs an intermediary matching layer that sits between exact string comparison and final bounding box assignment. This layer implements flexible matching logic that can handle non-unique tokens and partial matches, thereby improving adaptability while managing validation time through intelligent filtering and prioritization of match candidates.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent applies different matching strategies locally depending on the characteristics of the extracted information. For unique tokens, exact matching is used for speed; for non-unique or partial matches, more sophisticated comparison methods are applied. This local differentiation improves versatility without uniformly increasing validation time across all cases.

Inventive Principle:

Principle #3Local quality

3Measurement precision

If multiple OCR tokens are compared to find matching extracted information, then the accuracy improves, but the computational efficiency deteriorates

Engineering Contradiction:

Improveaccuracy of information locationVSAvoidefficiency of bounding box generation

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent segments both the extracted information and OCR data into discrete tokens, enabling efficient token-by-token comparison. This segmentation reduces the computational burden by breaking down complex matching tasks into smaller, manageable units, thereby maintaining accuracy while improving the efficiency of bounding box generation through incremental processing.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements a threshold-based filtering mechanism that stops comparison once sufficient matches are found, avoiding exhaustive comparison of all OCR tokens. This partial action approach maintains high accuracy by ensuring all relevant matches are captured while improving productivity by preventing unnecessary computational overhead from excessive comparisons.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20260017965A1Generating bounding boxes for information extracted from documents with large language models

Publication Date: 2026.01.15 INTUIT INC

AI summary

Certain aspects of the disclosure provide techniques for bounding box annotation for automated information extraction. A method generally includes obtaining an extracted field value for a field key of a document using optical character recognition (OCR) data generated based on the document, wherein: the OCR data comprises OCR tokens and bounding boxes, each associated with one OCR token; generating a value union bounding box surrounding value bounding box(es) of the bounding boxes, wherein: the value bounding box(es) satisfy a first threshold; and the value bounding box(es) are associated with first OCR token(s) of the plurality of OCR tokens that satisfy a second threshold when compared to one or more field value tokens of the extracted field value; and generating an output bounding box for display on a computing device with the document based on relative coordinates of the value union bounding box with respect to known dimensions of the document.