Document Layout Feature Extraction for Template Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computerized processing of business documents is error-prone due to variations in formatting, terminology, and information granularity, leading to inaccuracies in automated recognition and processing, especially when documents are received in image form with scanning differences and visual artifacts.
Innovation Solution
A computerized system employing a top-down Logical Layout Analysis (LLA) approach using an object recognizer to identify document objects, their location, size, and content, organizing this information into a one-dimensional vector array for classification, which allows for the generation of a 'document layout identifier' akin to a fingerprint, simplifying the grouping of documents by their layout structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If comprehensive feature extraction is performed on document objects including location, shape, and size, then classification accuracy improves, but system complexity increases
Solution Approach 1:
The patent segments the document analysis process into distinct stages: document imaging, object recognition, feature extraction, and classification. By dividing the comprehensive analysis into modular segments, the system manages complexity while maintaining accuracy through structured processing of location, shape, and size features at appropriate stages.
Solution Approach 2:
The patent transforms two-dimensional document layout information into one-dimensional feature vectors for classification. This dimensional transformation simplifies the data structure while preserving essential spatial relationships, enabling accurate classification without requiring complex multi-dimensional processing.
2Measurement precision
If multiple features are considered in classification, then document grouping accuracy improves, but processing time increases
Solution Approach 1:
The patent performs preliminary feature extraction during the document imaging and object recognition phases, preparing location, shape, and size features before classification. This preliminary preparation reduces the computational burden during the actual classification process, enabling accurate grouping without excessive processing time.
Solution Approach 2:
By converting multi-dimensional spatial features into one-dimensional feature vectors, the patent reduces the complexity of comparisons during classification. This dimensional reduction maintains grouping accuracy while significantly decreasing the time required for feature comparison and document grouping.
3Measurement precision
If detailed object recognition is performed, then data extraction accuracy improves, but computational resources increase
Solution Approach 1:
The patent segments object recognition into hierarchical levels, first identifying document-level structures and then extracting specific object features only where needed for classification. This segmented approach achieves detailed data extraction accuracy while avoiding unnecessary computational resources by processing only relevant features.
Solution Approach 2:
The patent extracts only the essential features (location, shape, size) needed for classification and data extraction, separating these from other document attributes. This selective extraction maintains high accuracy for classification purposes while minimizing computational resource consumption by ignoring non-essential features.
Data Source
AI summary
Image encoded documents are identified by recognizing known objects in each document with an object recognizer. The objects in each page are filtered to remove lower order objects. Known features in the objects are recognized by sequentially organizing each object in each filtered page into a one-dimensional array, where each object is positioned in a corresponding one-dimensional array as a function of location in the corresponding filtered page. The one-dimensional array is then compared to known arrays to classify the image document corresponding to the one-dimensional array.