Key-Value Matching Model for Document Image Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current key-value matching methods in document images suffer from low accuracy in detecting key-value pairs due to position detection issues, leading to unreliable matching results.
Innovation Solution
A method and apparatus for key-value matching that utilize a predetermined key-value matching model comprising a semantic segmentation submodel and an image matching submodel. This model identifies target attribute data and value regions with higher accuracy, enabling precise determination of matching relationships between attribute data and value data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If position detection methods are used to obtain position information about Keys and Values, then key-value matching can be performed, but the accuracy of detection results is low
Solution Approach 1:
The patent replaces traditional mechanical position detection methods with a deep learning-based key-value matching model that processes image data end-to-end. The model uses convolutional neural networks to directly identify and match key-value pairs without relying on separate position detection and character recognition steps, thereby improving both position detection accuracy and overall matching reliability
Solution Approach 2:
The key-value matching model is divided into multiple functional modules including an encoding module for extracting image features, a matching module for identifying key-value relationships, and a decoding module for outputting results. This segmentation allows each module to specialize in specific tasks, improving the overall accuracy of position detection and matching
Data Source
AI summary
A method, apparatus, readable medium and electronic device of key-value matching, the method inputs the image to be detected into a predetermined key-value matching model, to cause the predetermined key-value matching model to output a matching relationship between the attribute data and the attribute value data, in this way, it can not only provide an end-to-end network model for key-value matching, effectively improve the efficiency of key-value matching, but also obtain the target attribute value data region and the target attribute data region of higher accuracy by the semantic segmentation submodel in the predetermined key-value matching model, and then determine the matching relationship between the attribute data and the attribute value data in the image to be detected based on the target attribute data region and the target attribute value data region by the image matching submodel, thereby effectively improving the accuracy of the key-value matching result.