Nested Table Detection and OCR for Scanned Document Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to accurately extract text from nested tables in scanned documents with varying backgrounds and colors, leading to difficulties in detecting and extracting tabular data.
Innovation Solution
A two-stage process involving automated detection of nested tables, followed by cell detection and text extraction, using techniques like adaptive thresholding, contour detection, and optical character recognition (OCR) to generate editable file formats such as CSV or JSON.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional text extraction methods are used on nested tables with varying backgrounds and colors, then the extraction process is simple, but the extraction accuracy deteriorates significantly
Solution Approach 1:
The patent segments the nested table extraction process into distinct stages: outer table detection, nested table identification, cell detection, and text extraction. Each stage handles specific aspects of the complex extraction task, improving accuracy by breaking down the problematic nested structure with varying backgrounds into manageable components that can be processed systematically
Solution Approach 2:
The patent performs preliminary actions by first detecting the outer table boundaries and identifying nested table regions before attempting text extraction. This preliminary structuring of the document layout allows subsequent extraction steps to focus on specific regions with known context, thereby improving text extraction accuracy despite background variations
2Productivity
If automated detection methods are applied to nested tables, then productivity increases, but detection precision deteriorates due to complex structures
Solution Approach 1:
The detection process is segmented into specialized sub-detectors: outer table detector for boundary identification, nested table detector for identifying embedded tables, and cell detector for individual cell recognition. Each detector is optimized for its specific task, maintaining high precision while enabling automated processing of complex nested structures
Solution Approach 2:
The patent introduces intermediate processing steps between detection and extraction, including boundary refinement and region validation. These intermediary operations act as mediators that correct detection errors and refine table structure identification, thereby maintaining detection precision in automated high-speed processing
3Adaptability or versatility
If simple extraction methods are used, then the process is fast, but the ability to handle varying backgrounds and colors deteriorates
Solution Approach 1:
The patent applies local quality by using adaptive thresholding and region-specific processing parameters. Different background regions are analyzed with locally optimized settings, allowing the system to adapt to varying colors and patterns in different parts of the document without requiring uniform processing across the entire page, thus handling background variations effectively
Data Source
AI summary
Methods, systems, and apparatus are provided for detecting and extracting data from a nested table. A contour image of a first table disposed within a cell of a second table from an input image is identified. The first table on the input image is masked to generate a masked image of the second table that omits the first table. Cells of the first table and the second table are detected by identifying a plurality of regions of interest (ROIs), determining a row of a plurality of cells based on a common y-coordinate among of the plurality of ROIs, and generating cell parameter information for each of the plurality of cells that identifies the y-coordinate, a size, and a relative position of each determined cell. An output file is generated based on the cell parameter information of the plurality of cells.