Hybrid Image Tagging System Using Confidence Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated image recognition systems face challenges in accurately extracting information from images, with varying reliability and efficiency compared to text data, necessitating a combination of automated and manual tagging methods to improve accuracy and reduce costs.
Innovation Solution
A hybrid approach that combines automated image recognition with manual tagging, where automated recognition provides initial tags with a confidence measure, and if below a threshold, manual review is triggered, allowing for expert human input to upgrade tags, and utilizing an image processing system with I/O, automatic identification, and human reviewer logic to manage and prioritize image reviews.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated image recognition is used to tag images, then processing cost and time are reduced, but accuracy and reliability of tags decrease
Solution Approach 1:
The patent segments the image tagging process into two distinct stages: automated recognition for initial tag generation and manual review for accuracy verification. This segmentation allows the system to leverage the speed of automated systems while ensuring accuracy through selective human intervention, directly resolving the contradiction between processing speed and tagging accuracy.
Solution Approach 2:
The patent introduces manual reviewers as an intermediary layer between automated recognition and final tag output. This intermediary performs quality control by verifying and correcting automated tags, particularly for images with low confidence scores or complex content, thereby maintaining high accuracy while preserving the efficiency benefits of automated processing.
2Measurement precision
If manual tagging is performed for all images, then accuracy and reliability of tags improve, but processing cost and time increase
Solution Approach 1:
The patent applies partial action by having manual reviewers examine only a subset of images—specifically those with low confidence scores from automated recognition or complex content that requires human judgment. This selective approach maintains high tagging accuracy while avoiding the excessive time and cost costs of manual review for all images.
Solution Approach 2:
The patent changes the parameter of reviewer workload by dynamically adjusting which images require manual review based on confidence scores and image complexity. This parameter-based filtering ensures that manual resources are concentrated on images where they provide the most value, optimizing the balance between accuracy and processing speed.
3Productivity
If automated image recognition is performed first with confidence threshold filtering, then processing efficiency is improved, but images below threshold require additional manual review resources
Solution Approach 1:
The patent performs preliminary automated recognition and confidence scoring before routing images to manual review. This preliminary action filters out most images that don't require human review, reducing the burden on manual reviewers while maintaining processing efficiency. The system complexity is managed through automated routing logic that directs only necessary images to human reviewers.
Data Source
AI summary
An image recognition approach employs both computer generated and manual image reviews to generate image tags characterizing an image. The computer generated and manual image reviews can be performed sequentially or in parallel. The generated image tags may be provided to a requester in real-time, be used to select an advertisement, and/or be used as the basis of an internet search. In some embodiments generated image tags are used as a basis for an upgraded image review. A confidence of a computer generated image review may be used to determine whether or not to perform a manual image review.