Document Text Extraction Using Character Dimension Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for reading-order text extraction from documents fail to accurately account for character heights and widths, leading to improper characterization of line or paragraph breaks and misidentification of column breaks, resulting in incorrect extraction of text in reading order.
Innovation Solution
The Extractor scans a document to create bounding boxes for each string, identifying physical layout breaks and building a target document map to reorder strings according to a reading order, while considering character dimensions and frequencies, and uses a document tree to extract and tag semantic entities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional recursive procedures are used to segment documents by detecting prominent gaps in word positions, then the segmentation process is simple and fast, but character heights and widths are not taken into account, leading to misidentification of line breaks and column breaks
Solution Approach 1:
The patent changes the parameters used for segmentation by incorporating character heights and widths into the analysis. Instead of only detecting prominent gaps in word positions, the system now considers dimensional parameters of characters to distinguish between line breaks (where character positions change vertically) and column breaks (where character positions change horizontally), thereby improving measurement precision without sacrificing processing efficiency
Solution Approach 2:
The patent replaces the simple gap-detection mechanism with a more sophisticated analysis that substitutes dimensional measurement and spatial relationship evaluation. Rather than relying solely on horizontal gap detection, the system now uses character height and width parameters to mechanically distinguish between different types of breaks, improving identification accuracy while maintaining computational efficiency
2Device complexity
If conventional techniques cast paragraph or line breaks within a column as the beginning of a new segment or column, then the segmentation process is simplified, but text extraction in reading order becomes incorrect
Solution Approach 1:
The patent changes the segmentation parameters by incorporating character height and width measurements to distinguish between line breaks and column breaks. By analyzing the spatial dimensions of characters and their positional relationships, the system can reliably identify true column boundaries versus line endings within a column, ensuring accurate reading-order text extraction without significantly increasing algorithmic complexity
Solution Approach 2:
The patent introduces character dimensional parameters as intermediary measurements that mediate between simple gap detection and accurate break identification. These intermediary height and width measurements serve as additional criteria that help the system correctly differentiate between line breaks and column breaks, improving reliability while keeping the overall process manageable
3Productivity
If conventional segmentation methods do not account for word starting frequencies, break alignment, and relative straightness, then the segmentation process is faster, but the physical layout characterization becomes inaccurate
Solution Approach 1:
The patent enhances the segmentation parameters by incorporating word starting frequencies, break alignment, and relative straightness measurements. These additional parameters allow the system to accurately characterize physical layouts by detecting patterns in word positioning and break locations, while the efficient computational implementation maintains processing speed
Solution Approach 2:
The patent performs preliminary analysis of word starting frequencies, break alignment, and relative straightness before final segmentation. By pre-processing and analyzing these parameters in advance, the system can quickly identify patterns and make accurate segmentation decisions, preserving both processing speed and layout characterization accuracy
Data Source
AI summary
Methods and apparatus provide for an Extractor that receives a collection of strings and a bounding box(es) for each string. Each bounding box describes a position of at least a portion of a corresponding string in a source document. The source document includes multiple sections for presenting portions of the collection of strings in the source document. The Extractor arranges the collection of strings according to a reading order that corresponds to a language associated with the collection of strings. Upon arranging the collection of strings according to the reading order, the Extractor collects semantic entities from a target document that has the collection of strings ordered according to the reading order. For each collected semantic entity, the Extractor tags the collected semantic entity with a tag that describes a category of content that corresponds to a logical section of the target document from which the semantic entity was extracted.