OCR Confidence Verification via Embedding Space Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Optical Character Recognition (OCR) systems face challenges in accurately computing confidence scores for text recognition, leading to incorrect rejections or acceptances, especially in poor visibility conditions, and require access to internal recognition processes, making them unreliable and costly for applications like license plate recognition.
Innovation Solution
A method and system that compute confidence in OCR output by generating representations of text images and candidate strings in a common embedding space, using a projection matrix learned from annotated data, to compare similarity and provide a more reliable confidence score without relying on internal OCR processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If OCR systems use internal confidence scores to evaluate recognition reliability, then the system can automatically filter results, but the confidence scores are unreliable and lead to incorrect rejections or acceptances
Solution Approach 1:
The patent introduces an intermediary verification system that acts as a mediator between the OCR system and the final output. This verification system uses a different approach (comparing generated text representations with image representations in embedding space) to assess confidence, rather than relying on the OCR system's internal confidence scores. The intermediary verification mechanism provides more reliable confidence assessment while maintaining automatic processing capability.
2Measurement precision
If a second OCR system is used to verify results, then recognition accuracy may improve, but access to internal mechanics is not available and deficiencies cannot be overcome
Solution Approach 1:
The patent creates a simplified representation (copy) of the text image and text string in embedding space, which can be compared without accessing the internal mechanics of the OCR system. This copying approach allows verification of recognition accuracy while avoiding the complexity of accessing and comparing internal OCR processes. The embedding space representations serve as a simplified interface that enables verification without requiring knowledge of the OCR system's internal workings.
3Reliability
If human annotators are used for manual review, then verification reliability improves, but the process becomes costly and time consuming
Solution Approach 1:
The patent enables the system to verify its own output automatically through the embedding space comparison mechanism. The verification system serves itself by comparing generated text representations with image representations to assess confidence, eliminating the need for human annotators. This self-service capability maintains high verification reliability while dramatically reducing the time and cost associated with manual review processes.
Data Source
AI summary
A system and method for computing confidence in an output of a text recognition system includes performing character recognition on an input text image with a text recognition system to generate a candidate string of characters. A first representation is generated, based on the candidate string of characters, and a second representation is generated based on the input text image. A confidence in the candidate string of characters is computed based on a computed similarity between the first and second representations in a common embedding space.