Post-OCR Shape Clustering for Character Code Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optical character recognition (OCR) engines often produce errors in segmenting images and assigning character codes due to overlapping characters and poor image quality, leading to incorrect recognition of text in documents.
Innovation Solution
A post-OCR processing method that classifies clip images into clusters based on shape similarity, generates cluster images, and corrects erroneous OCR character codes by assigning new codes to clusters with shape differences, improving the accuracy of character recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If OCR engine processes documents with overlapping characters and poor image quality, then processing speed is maintained, but recognition accuracy deteriorates
Solution Approach 1:
The patent segments the document image into multiple clip images based on bounding boxes, then further segments each clip image into multiple super-vectors by dividing it into a grid of sub-images. This segmentation allows the system to process complex overlapping characters by analyzing smaller, more manageable portions independently, improving recognition accuracy without requiring complete reprocessing of the entire document.
Solution Approach 2:
The patent transforms 2D clip images into 1D super-vectors by extracting feature vectors from grid sub-images and concatenating them. This dimensionality transformation enables the system to represent complex character shapes in a compressed format that can be efficiently compared and clustered, allowing accurate recognition of overlapping characters without increasing processing complexity.
2Reliability
If OCR engine uses traditional character recognition methods, then processing time is short, but error rate increases due to overlapping characters
Solution Approach 1:
The patent performs preliminary actions by pre-dividing each clip image into a grid of sub-images and extracting super-vectors before the actual recognition process. This preliminary preparation allows the system to quickly compare and cluster characters during processing, reducing the time needed for accurate character recognition of overlapping characters without increasing overall processing time.
Solution Approach 2:
The patent creates a simplified representation (super-vector) of each clip image by extracting key features from grid sub-images. This copying approach allows the system to work with compressed representations rather than full images, enabling accurate character code assignment for overlapping characters while maintaining fast processing speeds.
3Measurement precision
If OCR engine processes documents with poor image quality, then output volume is maintained, but recognition precision deteriorates
Solution Approach 1:
The patent segments poor quality clip images into multiple grid sub-images, allowing the system to identify and process only the most informative portions. This segmentation enables precise character segmentation even in low-quality images by focusing analysis on specific regions rather than attempting to process the entire degraded image, maintaining processing throughput.
Solution Approach 2:
The patent changes the representation parameters by transforming clip images into super-vectors that capture essential character features. This parameter transformation allows the system to maintain high recognition precision for poor quality images by working with extracted features rather than raw pixel data, preserving processing speed while improving segmentation precision.
Data Source
AI summary
Systems, methods and computer program products on storage devices for shape clustering and applications in processing various documents, including an output of an optical character recognition (OCR) process. The output of an OCR process is classified into a plurality of clusters of clip images and a representative image for each cluster is generated to identify clusters whose clip images were incorrectly assigned character codes by the OCR process.