Document Layout Analysis for Label-Value Pair Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional OCR techniques extract text line-by-line, which is inefficient for documents with complex layouts, such as invoices, where labels and values are not properly ordered, leading to intermingled content and difficulty in processing.
Innovation Solution
A system that uses document layout analysis to group and order content into label-value pairs based on positional relationships, employing a binary tree structure to determine the reading order and ensure that labels and values are streamed together without intervening content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional OCR techniques extract text line-by-line from left to right, then the extraction process is simple and fast, but the extracted content is not in logical reading order for documents with complex layouts
Solution Approach 1:
The patent segments the document into multiple lines and groups them into blocks based on spatial relationships. Each line is extracted individually using OCR, then lines are grouped into blocks where lines within a block are ordered sequentially and blocks are ordered based on their positional relationships, reconstructing the logical reading order from the segmented lines
Solution Approach 2:
The patent transitions from one-dimensional line-by-line extraction to two-dimensional block-based ordering by considering both horizontal and vertical positions. Lines are first ordered within their blocks based on horizontal position, then blocks are ordered based on vertical position, adding a spatial dimension to the extraction process to achieve logical reading order
2Device complexity
If conventional OCR extracts text without considering layout, then the extraction process is simple, but labels and values become intermingled and difficult to process
Solution Approach 1:
The patent segments the document content into distinct lines and further groups lines into blocks based on spatial proximity and relationships. This segmentation allows the system to identify and extract label-value pairs as coherent units rather than intermingled text, improving data processing ease while maintaining reasonable process complexity
Solution Approach 2:
The patent introduces blocks as an intermediary structure between individual lines and the final extracted content. Blocks serve as intermediate groups that organize lines according to their spatial relationships, enabling the system to preserve the logical structure of label-value pairs while maintaining a systematic extraction process
3Loss of information
If the system uses document layout analysis to group and order content into label-value pairs, then the extracted content is in logical order, but the processing time and computational resources increase
Solution Approach 1:
The patent divides the document into lines and further into blocks based on spatial relationships, enabling efficient processing by working with smaller, organized units rather than the entire document at once. This segmentation reduces computational complexity while maintaining accurate content ordering through systematic block evaluation
Solution Approach 2:
The patent applies layout analysis selectively to groups of lines organized into blocks rather than analyzing every possible line combination. By focusing analysis on localized blocks and using spatial heuristics to determine reading order, the system achieves accurate content ordering with reduced processing time compared to exhaustive analysis
Data Source
AI summary
The embodiments disclosed herein relate to identifying phrases in an electronic document, where each token is one or more characters. Phrases are formed from the tokens, based on a position of each token relative to other tokens in the document. If the horizontal space between two tokens is less than a threshold, the two tokens are identified as a phrase. Information identifying phrases and tokens can be stored in a marked-up document. Value phrases can be identified by the content of the phrase. Thereafter, a label phrase can be identified based on proximity to the value phrase and/or the presence of an association symbol in the phrase. The label phrase and value phrase can be identified as a label-value pair, where the label identifies the type of content in the value phrase. A reading order of the document can be determined through the use of a binary tree.