Self-Attention Encoder Decoder for Document Reading Order
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing reading order detection methods rely on manually developed rules-based algorithms, which fail to provide satisfactory results for complex documents like tables and receipts, and require manual tuning for different languages, limiting their adaptability and accuracy.
Innovation Solution
A self-attention encoder/decoder structure is used to predict the reading order of document layout components by combining textual and visual information, allowing for the automatic learning of coherent sequences and adapting to variable document structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rule-based algorithms are used for reading order detection, then the system is simple to implement, but the accuracy and adaptability to complex documents deteriorates
Solution Approach 1:
The patent replaces manual rule-based algorithms with a neural network-based machine learning system. The neural network automatically learns reading order patterns from training data, substituting the mechanical rule-based approach with an adaptive learning system that achieves higher accuracy on complex documents while maintaining automated operation.
Solution Approach 2:
The system changes from fixed rule-based parameters to dynamic learned parameters. The neural network learns optimal reading order predictions by adjusting its internal parameters during training, allowing it to adapt to different document types and languages without manual rule modification.
2Use of energy by moving object
If rule-based algorithms are used for reading order detection, then the system requires minimal computational resources, but the adaptability to different languages and document types deteriorates
Solution Approach 1:
The neural network is designed as a universal reading order detection system that can handle multiple languages and document types through a single model. By training on diverse data including different languages and document formats, the system achieves multi-functionality without requiring separate rule sets for each case.
Solution Approach 2:
The system performs self-learning through automated training on labeled data, eliminating the need for manual rule creation and tuning. The neural network automatically adapts to different languages and document types by learning from training examples, providing self-service adaptability.
3Measurement precision
If manual tuning of rules is performed for different languages, then the reading order accuracy for specific languages improves, but the time and effort required deteriorates
Solution Approach 1:
The system performs preliminary learning during the training phase by exposing the neural network to labeled reading order data from multiple languages and document types. This preliminary action pre-configures the model with language-specific and document-type-specific knowledge, eliminating the need for time-consuming manual tuning when deploying to new languages.
Data Source
AI summary
A method including receiving an image representing a document including a plurality of layout components, identifying textual information associated with the plurality of layout components, identifying visual information associated with the plurality of layout components, combining the textual information with the visual information, and predicting a reading order of the plurality of layout components based on the combined textual information and visual information using a self-attention encoder/decoder.