Table Structure Determination Using Single Cell Tokens
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing table recognition technologies face challenges in accurately interpreting and processing tabular layouts within documents, particularly due to complex structures and nesting, which can lead to invalid table structures and difficulties in data extraction.
Innovation Solution
A computer-implemented method using a table recognition neural network that assigns a single token to each cell of a table, allowing for easier interpretation and correction of table structures, and converts the table structure into HTML or other markup languages for accurate data extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If complex table structures with nesting are used to represent real-world tables, then the representation capability is improved, but the accuracy of interpretation and processing deteriorates due to invalid structures and extraction difficulties
Solution Approach 1:
The patent segments the table recognition task into distinct components: detecting table boundaries, identifying row and column structures, and assigning cell types independently. This segmentation allows each component to be processed and validated separately, improving overall accuracy while maintaining the ability to represent complex nested structures through systematic breakdown of the table into manageable units.
2Productivity
If traditional table recognition methods are used, then processing of simple tables is sufficient, but accuracy deteriorates when dealing with complex structures and nesting
Solution Approach 1:
The patent implements feedback mechanisms where the detected table structure is validated against expected patterns and constraints. The system provides feedback loops that allow correction of detected structures, adjusting the interpretation based on consistency checks with the overall table layout and known table structure rules, thereby improving accuracy without sacrificing processing efficiency.
Solution Approach 2:
The patent performs preliminary detection and validation of table components before final structure assembly. By pre-identifying table boundaries, row/column intersections, and cell types in advance, the system prepares structured data that can be efficiently processed and validated, improving both accuracy and productivity through staged processing.
3Loss of information
If detailed table structures are extracted for accurate data interpretation, then data extraction capability is improved, but processing complexity and difficulty increase
Solution Approach 1:
The patent segments the table into standardized units (rows, columns, cells) with clearly defined properties and relationships. This segmentation enables systematic extraction of data while maintaining a manageable processing complexity through structured representation, where each cell's content and context can be independently accessed and processed.
Solution Approach 2:
The patent transforms the table structure into standardized parameters and attributes that can be systematically processed. By converting visual table elements into structured data parameters (cell positions, row/column indices, cell types), the system enables accurate data extraction while simplifying processing through parameter-based manipulation rather than complex structural analysis.
Data Source
AI summary
A computer implemented method of determining a table structure of a table. The method comprises receiving an image of the table. The method further comprises receiving the table structure of the table in response to inputting the image of the table into a table recognition neural network. The table structure comprises a single token assigned to each cell of the table. The single token assigned to each cell of the table is selected from a finite number of tokens. The table structure comprises a row of tokens for each row of the table.