Weak-to-Strong Tabular Annotation Alignment via OCR Tokens
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for generating bounding boxes for tabular data in documents require significant human intervention and are inefficient, leading to errors in aligning annotations within data tables.
Innovation Solution
An annotation alignment service with tabular data matching is used to automatically generate bounding boxes for tabular data by aligning character recognition tokens with weak annotations, utilizing a search algorithm to determine heuristic scores and coordinate alignments without human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human annotators manually generate bounding boxes for strong annotations, then annotation accuracy is improved, but time consumption and cost increase significantly
Solution Approach 1:
The system performs self-annotation by automatically generating bounding boxes using OCR token alignment with weak annotations. The annotation alignment service autonomously processes documents, matching OCR tokens to weak annotation data tables without human intervention, thereby achieving both accuracy and efficiency through automated self-service annotation.
Solution Approach 2:
The patent introduces an annotation alignment service as an intermediary between weak annotations (from ERP systems) and strong annotations (with bounding boxes). This service uses OCR token alignment and heuristic scoring to bridge the gap, converting weak annotations into accurate strong annotations automatically without requiring manual bounding box generation.
2Quantity of substance
If weak annotations without bounding boxes are used, then data quantity increases, but alignment accuracy with OCR tokens deteriorates
Solution Approach 1:
The patent replaces manual mechanical alignment processes with an automated computational system. The annotation alignment service uses OCR token matching, coordinate transformation, and heuristic scoring algorithms to automatically align weak annotations with OCR tokens, substituting manual precision-based methods with automated data-driven alignment.
Solution Approach 2:
The system transforms weak annotations into strong annotations by changing key parameters: adding bounding box coordinates, establishing OCR token alignments, and computing heuristic scores. This parameter transformation enables the use of large quantities of weakly annotated data while maintaining alignment accuracy through automated coordinate and token matching.
3Extent of automation
If existing weak-to-strong algorithms are used for tabular data, then automation is improved, but alignment accuracy deteriorates due to errors with repeated values and table boundaries
Solution Approach 1:
The patent segments the annotation alignment process into distinct components: OCR token processing, weak annotation parsing, coordinate transformation, heuristic scoring, and bounding box generation. This segmentation allows each component to be optimized independently, particularly for handling repeated values and table boundary detection, improving overall accuracy while maintaining automation.
Solution Approach 2:
The system implements feedback mechanisms through heuristic scoring that evaluates the quality of OCR token alignments with weak annotations. The scoring system provides feedback on alignment quality, allowing the system to adjust and refine bounding box assignments, particularly for repeated values and edge cases, thereby improving accuracy while maintaining full automation.
Data Source
AI summary
Systems and processes for aligning weakly-annotated tabular data to recognized characters in a document are provided. Data table annotations are grouped by column and value to produce a plurality of annotation groups. Character recognition tokens are received, and a search algorithm is performed to align the annotation groups to the tokens in a stepwise manner. A base column is designated, after which the search algorithm is performed on the base column annotation groups and the results are used to determine a vertical range of the data table. The search algorithm is then run for the other columns of the data table with the results being limited based on the vertical range. At each step, the annotations in an annotation groups are each aligned to one or more of the tokens. A bounding box is generated for each annotation in the data table and output to a target application.