OCR Word Box Merging for Positional Document Information Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information extraction systems face challenges in efficiently retrieving information from two-dimensional documents due to the loss of positional information during conversion to one-dimensional sequences and the need for extensive computational resources, and image processing models are complex and require significant training.
Innovation Solution
An information extraction system utilizing optical character recognition (OCR) tools to identify word boxes, assign BIO tags and labels, and merge word boxes with common labels to form combined words, reducing computational requirements and enhancing understanding of document content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If document information is extracted by converting two-dimensional documents to one-dimensional sequences, then information extraction can be performed, but positional information is lost
Solution Approach 1:
The patent applies dimensionality change by converting two-dimensional document layouts into one-dimensional sequences while preserving spatial relationships through coordinate encoding. Each word box retains its original (x, y) position information in the sequence, allowing the system to reconstruct the two-dimensional layout from the one-dimensional representation without requiring complex image processing models.
2Measurement precision
If image processing models are used for document information extraction, then extraction accuracy can be improved, but computational resources and training requirements increase significantly
Solution Approach 1:
The patent extracts only the necessary features from documents by using OCR to identify word boxes and their positions, then processes this extracted information through simple sequence operations. This avoids the computational overhead of training and running complex image processing models while maintaining extraction accuracy through targeted feature selection and BIO tagging.
Solution Approach 2:
The patent employs lightweight, disposable processing steps instead of expensive, long-lived trained models. The system uses simple sequence processing, BIO tag assignment, and coordinate-based position reconstruction that require minimal computational resources and no extensive training, replacing the need for heavy image processing models.
3Ease of operation
If word boxes are processed individually, then processing simplicity is maintained, but information about related words is lost
Solution Approach 1:
The patent merges individual word box processing with semantic relationship detection by grouping word boxes that share the same BIO tags and labels. Words with matching labels are combined into unified representations that preserve both individual word information and their semantic relationships, achieving both processing simplicity and information retention.
Data Source
AI summary
Embodiments are described for a system comprising a memory and at least one processor coupled to the memory. The at least one processor is configured to receive optical character recognition (OCR) information of a document and determine a beginning, inside, and outside (BIO) tags and labels of the one or more word boxes based on the OCR information. The at least one processor is further configured to group a first word box and a second word box based on BIO tags of the first and the second word boxes and merge the first and the second word boxes into a combined word box based on a label of the first word box matching a label of the second word box. Finally, the at least one processor is configured to output the combined word box and the label of the first word box.