Document Image Consensus Reconstruction for Annotation Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document scanning technologies face difficulties in separating variable text from fixed text in filled-out forms, especially when the form layout information is not available, leading to challenges in processing and reconstructing clean documents from annotated images.
Innovation Solution
A computer-implemented method and system that performs global and local alignments of annotated document images to form a consensus document image, extracts annotation data by comparing annotated images with the clean document, and uses a hierarchical search to correct distortions and extract annotation pixels, allowing for rotation, translation, and scaling adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If automated methods are used to separate annotated data from document images without pre-existing form layout information, then the extent of automation is improved, but the difficulty of detecting and measuring increases
Solution Approach 1:
The system performs preliminary actions by selecting a representative annotated document image from the set before processing begins. This representative image serves as a reference template that guides the subsequent alignment and separation processes, enabling automated processing without requiring pre-existing form layout information.
Solution Approach 2:
The system creates a consensus document image by aligning and combining multiple annotated document images. This consensus image acts as a synthesized template that captures the common structure across all input images, enabling the system to automatically identify and separate variable annotated data from fixed form elements without requiring prior knowledge of the form layout.
2Measurement precision
If multiple annotated document images are aligned and processed to form a consensus document image, then the accuracy of annotation extraction is improved, but the processing time and computational resources increase
Solution Approach 1:
The system segments the processing task into distinct phases: selecting a representative image, performing global alignment to correct overall positioning and orientation, performing local alignment to refine positioning of specific regions, and generating the consensus document image. This segmentation allows the system to process multiple images efficiently by breaking down the complex task into manageable steps that can be optimized independently.
3Manufacturing precision
If global and local alignment transformations are applied to correct distortions in annotated document images, then the manufacturing precision of the reconstructed document is improved, but the device complexity increases
Solution Approach 1:
The system applies different levels of alignment transformations to different aspects of the document images. Global alignment handles overall positioning, rotation, and scaling of entire images, while local alignment refines the positioning of specific regions and features. This local quality approach allows the system to achieve high precision in document reconstruction by applying appropriate transformation complexity only where needed, rather than uniformly across all images.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method and system for reconstructing a clean document from annotated document images and/or extracting annotations therefrom are provided. The method includes receiving (S200) a set of at least two annotated document images into computer memory, selecting (S300) a representative image from the set of annotated document images, performing (S400) a global alignment on each of the set of annotated document images with respect to the selected representative image, and forming (S600) a consensus document image based at least on the aligned annotated document images. A clean document based at least on the consensus document image is then formed which can be used for extracting the annotations.