Bill Image Recognition via Dynamic Key-Value Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing bill image recognition technologies based on fixed templates have poor scalability and versatility, leading to low recognition accuracy due to inconsistent layouts and different bill structures.
Innovation Solution
A method and apparatus that perform text detection on bill images, determine attribute and relationship information sets for text boxes, and infer the type and structural relationships between them, allowing for the extraction of structured bill data without relying on a fixed template.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If template matching is used for bill information extraction, then extraction speed is improved, but accuracy deteriorates due to inconsistent layouts and different bill structures
Solution Approach 1:
The patent changes the fundamental parameter of bill recognition from fixed template matching to dynamic key-value pair extraction. By transforming the recognition approach to identify semantic relationships between keys and values regardless of position, the system maintains high extraction speed while achieving accurate results across varied bill layouts and structures.
Solution Approach 2:
The patent creates a universal recognition framework that can handle multiple bill types and layouts through a single key-value pair extraction mechanism. This multi-functional approach eliminates the need for separate templates for different bill formats, enabling the system to accurately process diverse bill structures while maintaining consistent extraction performance.
2Measurement precision
If fixed templates are used for each bill format, then recognition accuracy is improved for specific formats, but device complexity and maintenance difficulty increase
Solution Approach 1:
The patent implements a universal key-value pair extraction framework that serves all bill formats through a single mechanism. This eliminates the need to maintain multiple format-specific templates, reducing system complexity while preserving the ability to accurately recognize different bill types through semantic relationship identification.
Solution Approach 2:
The system automatically identifies key-value relationships based on semantic understanding without requiring manual template configuration for each bill format. This self-adapting approach reduces maintenance burden and complexity while maintaining high recognition accuracy across various bill structures.
3Ease of operation
If fixed templates are used for bill recognition, then extraction process is simplified, but adaptability and versatility deteriorate
Solution Approach 1:
The patent transforms the extraction approach from rigid template-based matching to flexible key-value pair identification. This parameter change enables the system to maintain simple extraction operations while simultaneously achieving high adaptability to different bill formats through semantic relationship recognition between keys and values.
Solution Approach 2:
The system dynamically identifies key-value relationships based on the actual bill content and structure rather than relying on fixed positional templates. This dynamic approach simplifies the extraction process while enhancing versatility, allowing the system to adapt to various bill formats automatically through semantic understanding.
Data Source
AI summary
A method, apparatus, device and storage medium for recognizing a bill image may include: performing text detection on a bill image, and determining an attribute information set and a relationship information set of each text box of at least two text boxes in the bill image; determining a type of the text box and an associated text box that has a structural relationship with the text box based on the attribute information set and the relationship information set of the text box; and extracting structured bill data of the bill image, based on the type of the text box and the associated text box that has the structural relationship with the text box.