Confidence Filtering for OCR and Classification Reassessment

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image recognition techniques face challenges in accurately determining confidence metrics for text recognition and classification models, particularly with complex image formats, low resolution images, and multilingual content, leading to unreliable and time-consuming manual audits.

Innovation Solution

A filter circuitry that combines confidence metrics from text recognition and classification models to determine a global confidence metric, flagging images requiring additional review and reducing data transmission and human intervention by only reassessing flagged images.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual audits are performed to verify text recognition and classification accuracy, then measurement precision is improved, but loss of time and productivity deteriorate

Engineering Contradiction:

Improveconfidence metric accuracyVSAvoidmanual audit time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs self-verification by automatically generating and analyzing confidence metrics for text recognition and classification results. The filter circuitry autonomously identifies low-confidence predictions and triggers reprocessing without human intervention, allowing the system to serve its own verification needs and eliminate manual audit requirements.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system implements a feedback mechanism where confidence metrics from initial processing are evaluated, and low-confidence results are automatically flagged for reprocessing. This closed-loop feedback system continuously improves accuracy by iteratively refining predictions based on confidence threshold evaluation, replacing manual verification with automated feedback-driven correction.

Inventive Principle:

Principle #23Feedback

2Reliability

If all images are reassessed to ensure accuracy, then reliability is improved, but loss of energy and processing requirements worsen

Engineering Contradiction:

Improveresult accuracyVSAvoidprocessing energy

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system applies different processing quality levels to different images based on their confidence metrics. High-confidence images receive minimal or no reprocessing, while low-confidence images undergo intensive reevaluation. This localized quality approach ensures reliability for critical cases while conserving processing energy for the majority of confident predictions.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

Instead of uniformly reassessing all images, the system performs partial reprocessing only on images that fall below confidence thresholds. This selective application of reprocessing action ensures sufficient reliability for uncertain cases while avoiding excessive processing energy consumption on already confident predictions.

Inventive Principle:

Principle #16Partial or excessive action

3Measurement precision

If confidence metrics are determined for all predicted characters and classifications, then measurement precision is improved, but device complexity and processing requirements worsen

Engineering Contradiction:

Improveconfidence metric determinationVSAvoidfilter circuitry complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system extracts only the necessary confidence metric information from text recognition and classification models, rather than processing all possible attributes. The filter circuitry isolates and evaluates only the confidence scores associated with predicted characters and classifications, separating essential reliability data from unnecessary complexity in the overall system architecture.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS20250371897A1Methods, systems, articles of manufacture, and apparatus to determine confidence metrics associated with text recognition models and classification models

Publication Date: 2025.12.04 NIELSEN CONSUMER LLC

AI summary

Methods, systems, articles of manufacture, and apparatus to determine confidence metrics associated with text recognition models and classification models are disclosed. An example apparatus comprises interface circuitry, machine-readable instructions, and at least one processor circuit to be programmed by the machine-readable instructions to cause a text recognition model to predict characters in an image, and determine first confidence metrics associated with sets of the predicted characters, cause a classification model to classify the sets of the predicted characters by determining predicted classifications for the sets of the predicted characters, and determine second confidence metrics associated with the predicted classifications, determine third confidence metrics based on the first confidence metrics and the second confidence metrics, compare the third confidence metrics to a threshold, and in response to the third confidence metrics satisfying the threshold, prevent a transmission of the image to a database.