Semantic Unit Identification in PDF Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying semantic units in electronic documents described in page description languages, such as PDF, fail to accurately reconstruct the logical order of text and group characters into words, especially in complex documents with multiple columns or arbitrary output ordering schemes, leading to incorrect processing and identification of spaced-out characters.
Innovation Solution
A method that identifies words by creating geometric strips and zones from glyphs, using geometric and font-related parameters to determine potential boundaries between semantic units, and sorting them based on vertical and horizontal positions, while combining adjacent units and eliminating redundant text.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If text fragments are extracted by simply replaying page description language instructions, then the visual rendering is preserved, but the logical structure of the text is lost
Solution Approach 1:
The patent segments the text extraction process into multiple stages: first extracting text fragments with their geometric properties, then analyzing spatial relationships to reconstruct logical order, and finally identifying semantic units. This segmentation allows preservation of visual fidelity while recovering logical structure through systematic analysis of fragment positions and relationships.
Solution Approach 2:
The patent performs preliminary analysis of geometric properties and spatial relationships between text fragments before final semantic unit identification. By pre-calculating distances, positions, and geometric characteristics, the system prepares data structures that enable accurate reconstruction of logical text order without losing visual rendering information.
2Ease of manufacture
If text fragments are ordered according to page description language sequence, then the extraction process is simple, but the logical reading order is incorrect in multi-column layouts
Solution Approach 1:
The patent dynamically determines text fragment ordering based on their geometric positions and spatial relationships rather than relying on fixed page description language sequence. The system adapts the extraction order according to detected column structures, line positions, and fragment coordinates, enabling accurate logical order identification in complex multi-column layouts while maintaining simple extraction processing.
Solution Approach 2:
The patent transitions from one-dimensional sequential extraction (following page description language instructions) to two-dimensional spatial analysis by considering both x and y coordinates of text fragments. This dimensional approach enables correct identification of logical reading order in multi-column documents by analyzing the geometric arrangement of fragments across the page.
3Measurement precision
If words are identified by explicit marks, then identification is accurate, but page description languages do not provide such marks
Solution Approach 1:
The patent replaces the mechanical approach of relying on explicit word boundary marks with a geometric analysis system that infers word boundaries through spatial relationships between characters. By calculating distances between character centroids, analyzing font metrics, and detecting spacing patterns, the system substitutes the need for explicit marks with computational geometric analysis, achieving accurate word identification without additional marking complexity.
Solution Approach 2:
The patent introduces geometric properties and spatial relationships as intermediary elements between the raw text fragments and the final word identification. These geometric intermediaries (positions, distances, orientations) serve as mediators that enable accurate word boundary detection without requiring explicit word marks in the source document.
4Reliability
If characters are grouped into semantic units using geometric analysis, then logical order is reconstructed, but processing time increases
Solution Approach 1:
The patent applies partial geometric analysis focused on the most critical aspects: calculating only the necessary spatial relationships between characters, analyzing font metrics selectively, and detecting word boundaries at key positions. This partial action approach maintains reliable logical structure reconstruction while reducing unnecessary computational overhead and processing time.
Data Source
AI summary
A method of identifying semantic units in an electronic document includes the steps of: providing an electronic document being described in a page description language, the document having at least one page having a plurality of text fragments, each text fragment including a plurality of glyphs that have not been identified as semantic units, the document further including geometric information and page description language parameters; determining strips of at least one glyph by comparing the geometric position of subsequent glyphs; determining zones of at least one strip wherein a zone is defined by the combined area of strips, the geometrical areas of which overlap with each other; determining a boundary between two semantic units in a zone based on the geometric properties of the glyphs; sorting the identified semantic units in the zone in a sorted list; and, combining subsequent semantic units in the sorted list according to geometric considerations.