Glyph Model Clustering for OCR Speed and Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Optical Character Recognition (OCR) systems face high error rates and slow processing speeds, making them unsuitable for efficient integration with JBIG2 compression techniques, which also suffer from high error rates and slow processing.
Innovation Solution
Implementing JBIG2-type matching techniques for error correction and speeding up OCR processing by classifying glyphs into model classes, using perceptually lossless matchers, and generating a model reference page for efficient OCR processing across multiple pages, allowing for backward and forward propagation of ASCII values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional OCR feature extraction methods are used to identify characters by analyzing shape features against rule sets, then character recognition can be performed, but processing speed is slow and error rates are high
Solution Approach 1:
The patent applies JBIG2 pattern matching techniques as a preliminary action before OCR processing to pre-identify and group similar glyphs. By performing shape-based clustering in advance using perceptually lossless matchers, the system reduces the search space for OCR, thereby improving both speed and accuracy without requiring full OCR processing for every glyph
Solution Approach 2:
The patent introduces an intermediary JBIG2 matching layer between the input image and OCR processing. This intermediary performs initial glyph grouping and similarity assessment, acting as a mediator that filters and organizes glyphs before they undergo expensive OCR feature extraction, thus improving overall system efficiency
2Reliability
If JBIG2 pattern matching techniques are applied to improve OCR results through post-processing, then error rates decrease for low confidence glyphs, but processing time increases
Solution Approach 1:
Instead of performing JBIG2 pattern matching as a post-processing step, the patent applies it as a preliminary action before OCR. By pre-grouping glyphs into model classes based on shape similarity, the system eliminates the need for time-consuming post-processing comparisons, thereby maintaining accuracy improvements without the time penalty
Solution Approach 2:
The patent inverts the conventional sequence by applying JBIG2 matching before OCR rather than after. This reversal allows the system to benefit from accurate glyph grouping while avoiding the time loss associated with post-OCR processing, as the matching operates on raw glyphs rather than processed OCR results
3Measurement precision
If OCR processing is applied to every glyph individually to ensure accurate ASCII value assignment, then recognition accuracy is maintained, but processing speed decreases significantly
Solution Approach 1:
The patent segments glyphs into model classes based on shape similarity using JBIG2 matching. By dividing the set of all glyphs into distinct model groups, the system can assign ASCII values to model representatives rather than processing every individual glyph, thereby maintaining accuracy while dramatically improving throughput
Solution Approach 2:
The patent creates model reference pages that serve as copies or representatives for groups of similar glyphs. Instead of processing each glyph individually, the system processes the model reference page once and propagates the results to all matching glyphs, achieving both accuracy and speed through this copying approach
4Productivity
If model reference pages are generated and OCR processing is applied only once per model class, then processing speed improves significantly, but system complexity increases
Solution Approach 1:
The patent creates model reference pages that serve multiple functions: they act as compression dictionaries for JBIG2, as training data for OCR, and as templates for glyph matching. This multi-functionality justifies the added complexity by eliminating the need for separate processing systems for each function
Solution Approach 2:
The patent merges the JBIG2 compression process with the OCR processing pipeline by integrating model class generation into both workflows. By combining these functions into a unified process, the system achieves speed improvements without proportionally increasing complexity, as the same data structures serve dual purposes
Data Source
AI summary
A system and method may include a processor that groups the glyphs of a document into font character models. OCR processing may be performed to identify the ASCII value of the font character models, with the results mapped to the glyphs contained with those models, thereby identifying the text of the original document. This results in fewer calls to an OCR engine, thereby providing a significant speedup. Further, when a model is assigned differing text values by the OCR engine, the system and method may identify the value most likely to be correct, thereby improving the accuracy of the output text.