Table Structure Recognition Using Separation Line Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing table structure recognition technologies struggle to accurately identify the organizational structure of tables in images, particularly those with complex structures, geometric deformations, or bending, and often fail to recognize tables without clear boundaries.
Innovation Solution
A deep learning-based approach that determines reference points on separation lines within a table image using feature maps, predicting separation lines to accurately identify the structure of tables, including those with unbounded cells and large blank areas, by employing a feature extraction module, separation line prediction modules, and a cell processing module.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing table structure recognition technologies are used, then simple tables can be recognized, but complex tables with geometric deformations or bending cannot be accurately identified
Solution Approach 1:
The patent segments the table structure recognition task into multiple components: detecting separation lines first, then identifying cells based on these lines. This segmentation allows the system to handle complex tables by breaking them down into manageable parts, improving both adaptability and precision simultaneously
Solution Approach 2:
The patent employs parameter changes by using different feature extraction parameters and detection thresholds to adapt to various table structures. The system adjusts its detection parameters based on the characteristics of the input image, enabling accurate recognition of both simple and complex tables including those with geometric deformations
2Adaptability or versatility
If existing table structure recognition technologies are used, then tables with clear boundaries can be identified, but tables without clear boundaries or with large blank areas fail to be recognized
Solution Approach 1:
The patent performs preliminary action by first detecting separation lines and establishing a structural framework before identifying individual cells. This preliminary structuring enables the system to reliably identify cells even in tables without clear boundaries or with large blank areas, as the separation lines provide the necessary structural context
Solution Approach 2:
The patent introduces separation lines as an intermediary element that mediates between the image input and cell identification. These separation lines serve as reliable markers that enable consistent cell layout identification across diverse table types, including those with ambiguous boundaries
3Measurement precision
If deep learning-based approaches with feature maps are employed, then recognition accuracy for complex tables improves, but computational complexity increases
Solution Approach 1:
The patent segments the deep learning process into distinct modules: feature extraction, separation line detection, and cell identification. This modular segmentation reduces system complexity by making each component independent and manageable, while maintaining high overall accuracy through the coordinated operation of these specialized modules
Data Source
AI summary
According to implementations of the present disclosure, a solution of table structure recognition is provided. A first set of reference points in the image including the table is determined based on a first feature map. The first feature map is generated from an image, and the first set of reference points are candidate points on the separation lines of a first type of the table. Based on at least a part of the first feature map and features of the first set of reference points, a set of predicted separation lines of the first type of the table can be determined in the image. The structure of the table is determined based at least on the set of predicted separation lines of the first type. In this way, the tables of various structures can be restored from the image.