Document Classification via Scan Line Signal Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional document classification techniques are inefficient for documents with sparse text and multi-page documents, as they are either time-consuming, resource-intensive, or fail to handle variable page counts effectively.
Innovation Solution
A method using signal processing that generates a signal representation of documents based on non-white pixels in scan lines, allowing for comparison with previously classified documents using dynamic time warping to determine classification, thereby overcoming the limitations of text-based and image-based classification models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image-based classification models (CNN) are used to classify documents, then classification accuracy is improved, but processing resources and memory requirements increase significantly
Solution Approach 1:
The patent extracts only the essential visual features from documents (black pixel positions and densities on scan lines) rather than processing entire document images. This extraction approach maintains classification accuracy by preserving layout information while dramatically reducing the data volume that requires processing, thus lowering computational resource requirements.
Solution Approach 2:
Instead of using complex image-based classification models that require extensive training data and processing power, the patent creates simplified signal representations (copies of essential document features) that can be efficiently compared using dynamic time warping. These signal copies retain the discriminatory power needed for classification while being computationally lightweight.
2Adaptability or versatility
If image-based classification models are used to handle multi-page documents, then classification capability is improved, but the models become too large and require extraordinary processing resources
Solution Approach 1:
The patent segments multi-page documents into individual page representations, where each page is converted to a signal representation independently. These segmented representations are then processed separately through dynamic time warping comparisons, allowing the system to handle variable page counts without requiring a single oversized model that could process arbitrarily long sequences.
Solution Approach 2:
The patent employs dynamic time warping, a dynamic programming technique that can flexibly align and compare signal representations of documents with varying page counts. This dynamic approach allows the classification system to adapt to different document lengths without requiring a fixed-size model, maintaining versatility while keeping computational complexity manageable.
3Productivity
If model-based classification is used for documents with sparse text, then text processing speed is improved, but classification accuracy deteriorates due to inability to recognize layout-based documents
Solution Approach 1:
The patent replaces text-based mechanical classification approaches with signal processing methods that analyze visual layout patterns. By converting document layouts into signal representations based on black pixel positions and densities, the system can classify documents with sparse text or no text at all, achieving both speed and accuracy for layout-based classification.
Data Source
AI summary
Aspects of the present disclosure provide techniques for document classification through signal processing. Embodiments include receiving a document for classification. Embodiments include generating an image of the document. Embodiments include producing a signal representation of the document based on numbers of non-white pixels in each horizontal scan line or vertical scan line of the image of the document. Embodiments include comparing the signal representation of the document to signal representations of previously-classified documents. Embodiments include determining, based on the comparing, a classification for the document. Embodiments include performing additional processing with respect to the document based on the classification for the document.