Document Image Region Classification for OCR Quality Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
OCR systems struggle to accurately extract data from documents of varying quality without providing insights into their accuracy, leading to inconsistent results due to factors like poor resolution and noise.
Innovation Solution
A method and system for determining document quality using machine learning models to segment images into regions, classify them into quality classes, and compute a cumulative quality score based on weighted averages of region classifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If OCR is performed on documents of varying quality, then data extraction can be attempted, but extraction accuracy deteriorates due to poor resolution and noise
Solution Approach 1:
The patent segments the document image into multiple regions and evaluates each region separately using quality metrics. This allows the system to identify high-quality regions for reliable OCR and low-quality regions that should be excluded or processed differently, thereby maintaining extraction accuracy while preserving productivity.
Solution Approach 2:
The patent performs quality assessment before OCR processing by evaluating regions based on multiple metrics (resolution, noise, text clarity). This preliminary action enables the system to determine which regions are suitable for OCR, preventing inaccurate extraction from poor-quality regions while maintaining overall extraction capability.
2Measurement precision
If multiple quality metrics are evaluated to improve assessment accuracy, then reliability of quality determination improves, but system complexity increases
Solution Approach 1:
The patent divides the quality assessment into multiple independent metric evaluations (resolution, noise, text clarity) that can be computed separately and then combined. This segmentation allows each metric to be optimized independently while maintaining overall system manageability and accuracy.
Solution Approach 2:
The patent creates a unified quality assessment system that integrates multiple different metrics into a single comprehensive evaluation framework. This multi-functional approach allows the system to assess various aspects of document quality through a single integrated process, improving measurement precision without proportionally increasing complexity.
3Reliability
If regions are classified into multiple quality classes, then extraction reliability improves, but processing time increases
Solution Approach 1:
The patent performs region classification into quality classes before OCR processing. This preliminary categorization allows the system to quickly identify and prioritize high-quality regions for extraction, reducing the time spent on low-quality regions while maintaining high extraction reliability through selective processing.
Solution Approach 2:
The patent applies different processing strategies to different quality classes of regions. High-quality regions undergo full OCR processing, while low-quality regions are either excluded or processed with simplified methods. This local differentiation maintains extraction reliability for suitable regions while reducing overall processing time.
Data Source
AI summary
A method and system of determining quality of a document image is disclosed that includes segmenting, by one or more processors, a document image into a plurality of regions each of which comprises text data. The plurality of regions is classified into one of a plurality of image quality classes based on a determination of a highest prediction value from one of a plurality of machine learning models. The plurality of machine learning models is trained corresponding to one of the plurality of image quality classes. A cumulative quality score for the image is computed based on a weighted average of a number of regions classified into each of the plurality of image quality classes. The quality of the image is determined based on the cumulative quality score.