Document Field Detection with Visual Words and Spatial Statistics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face challenges in automatically extracting information from documents with complex layouts due to varied positions of text, images, and tables, and ambiguous boundaries, often requiring large labeled datasets that are not feasible for real-life tasks.
Innovation Solution
A method for generating and optimizing a codebook of visual words using local descriptors and mutual information maximization to detect fields in document images, which relies on the spatial structure of documents and does not require extensive pre-labeled data, utilizing morphological preprocessing, keypoint extraction, and clustering to create a visual codebook.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional image processing methods are used for document analysis, then the system can handle complex document layouts, but it requires large labeled datasets which are not feasible for real-life tasks
Solution Approach 1:
The patent applies preliminary action by pre-processing document images through morphological operations (erosion, dilation, opening, closing) and keypoint detection (MSER, FAST, ORB, SIFT) before field position prediction. This preliminary structuring of the data enables the system to learn from fewer labeled examples by extracting meaningful features in advance, reducing the need for large labeled datasets while maintaining prediction accuracy
2Adaptability or versatility
If codebook-based visual word methods are used, then the system can extract features from document images, but it struggles with ambiguous boundaries and varied positions of document elements
Solution Approach 1:
The patent transitions from traditional 2D image processing to 3D spatial reasoning by modeling document fields as three-dimensional objects with width, height, and position coordinates. The field position prediction network outputs predictions in three dimensions (horizontal position, vertical position, and field width), enabling the system to handle varied layouts while maintaining precise boundary detection through volumetric spatial understanding
Data Source
AI summary
A method of detecting fields in document images includes: receiving a codebook comprising a set of visual words, each visual word corresponding to a center of a cluster of local descriptors; calculating, based on a set of user labeled document images, for each visual word of the codebook, a respective frequency distribution of a field position of a specified labeled field with respect to the visual word; loading a document image for extraction of target fields; calculating a statistical predicate of a possible position of a target field in the document image based on the frequency distributions; and detecting, using the trained model, fields in the document image based on the calculated statistical predicate.