Merging Unverified Entity Identities in Media Libraries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Managing media libraries often results in a large number of unverified entity identities, which can be cumbersome and inefficient, requiring tedious classification by users.
Innovation Solution
The method involves detecting faces in media content and grouping similar feature vectors into clusters, using metadata correlations such as geographical, temporal, and social group associations to merge unverified identities into verified ones, reducing the cognitive burden on users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face detection and clustering is performed on all media content, then the number of detected entity identities increases, but the number of unverified identities increases making the library unmanageable
Solution Approach 1:
The patent merges multiple feature vector clusters that represent the same entity by analyzing metadata correlations. When two clusters are found to be correlated through shared metadata (geographical location, time period, social groups), their feature vectors are combined into a single unified cluster, reducing the total number of unverified identities while maintaining detection accuracy.
Solution Approach 2:
Metadata acts as an intermediary to connect and verify entity identities across different media content. By using metadata correlations (geographical, temporal, social) as a bridge, the system can indirectly verify that different feature vector clusters represent the same entity without requiring direct face recognition between all pairs of images.
2Reliability
If manual classification of unverified identities is required, then verification accuracy can be maintained, but user cognitive burden and time consumption increase significantly
Solution Approach 1:
The system performs preliminary automated merging of feature vector clusters based on metadata correlations before presenting results to the user. This preliminary action pre-processes and reduces the number of identities that require manual verification, maintaining reliability while significantly reducing the time and cognitive burden on users.
Solution Approach 2:
The system uses metadata correlations as feedback to automatically verify and merge entity identities. By continuously analyzing metadata relationships and updating cluster assignments based on this feedback, the system maintains high verification reliability while automating the process and reducing user intervention requirements.
3Productivity
If feature vector clusters are merged using liberal confidence thresholds, then the number of unverified identities decreases, but the risk of incorrect merging increases
Solution Approach 1:
Instead of relying solely on face similarity in feature space, the patent introduces metadata correlation as an additional dimension for verifying identity merges. By combining face vector distance metrics with metadata correlation strength across multiple dimensions (geography, time, social groups), the system can use more liberal confidence thresholds while maintaining accuracy through multi-dimensional verification.
Data Source
AI summary
Systems, methods, and computer-readable media for reducing a number of unverified persons detected in media content are provided.