Face Track Recognition Using Non-Linear Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition in video delivery is a computationally intensive process that requires user input and often results in time-consuming and inaccurate labeling due to confusing user interfaces and incorrect labeling.
Innovation Solution
A method that determines face tracks in videos by receiving labels from users, extracting features, correlating them using a non-linear weighting function to generate confidence scores, and applying these scores to identify unlabeled face tracks, thereby reducing the need for extensive user input and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face recognition is performed using traditional methods with extensive user input, then labeling accuracy can be improved, but the time required and user effort increase significantly
Solution Approach 1:
The system applies partial action by requiring users to label only a subset of face tracks (e.g., 20%) rather than all face tracks. The remaining unlabeled tracks are automatically processed using the learned weighting function and feature correlation, significantly reducing user time investment while maintaining high labeling accuracy through the combination of partial manual input and automated inference.
Solution Approach 2:
The system performs self-service by automatically generating feature weights and correlating features between labeled and unlabeled face tracks using the weighting function. This automated process eliminates the need for users to manually review and verify all face tracks, allowing the system to independently complete the recognition task for unlabeled tracks based on the patterns learned from partially labeled data.
2Measurement precision
If traditional face recognition methods are used, then comprehensive labeling can be achieved, but the user interface becomes confusing and labeling becomes incorrect
Solution Approach 1:
The system segments the face tracking and labeling task into two distinct parts: (1) a simplified user interface for providing initial labels on a small subset of face tracks, and (2) an automated processing pipeline that handles the remaining tracks. This segmentation reduces the complexity of the user interface by limiting user interaction to only the most critical initial labeling task, while the automated system handles the bulk of the work.
Solution Approach 2:
The system introduces an intermediary automated processing layer between user input and final labeling output. This intermediary component (the weighting function and feature correlation mechanism) processes the limited user-provided labels and automatically generates labels for the remaining tracks, acting as a mediator that translates simple user input into comprehensive accurate labeling without requiring users to navigate complex interfaces.
3Measurement precision
If all face tracks require user labeling, then complete accuracy is achieved, but computational intensity and processing time increase
Solution Approach 1:
The system applies partial action by processing only a subset of face tracks through the computationally intensive automated feature correlation and weighting functions, rather than all tracks. The user provides labels for a partial set (e.g., 20%) of tracks, and the system uses these partial labels to train the weighting function, which then efficiently processes the remaining tracks without requiring full computational intensity for all tracks.
Solution Approach 2:
The system performs preliminary action by first processing a small subset of face tracks to learn and train the weighting function and feature correlation models. This preliminary processing establishes the computational patterns and weights needed for rapid automated processing of the remaining tracks, significantly reducing the computational intensity required for the bulk of the labeling task.
Data Source
AI summary
In one embodiment, a method determines known features for existing face tracks that have identity labels and builds a database using these features. The face tracks may have multiple different views of a face. Multiple features from the multiple faces may be taken to build the face models. For an unlabeled face track without identity information, the method determines its sampled features and finds labeled nearest neighbor features with respect to multiple feature spaces from the face models. For each face in the unlabeled face track, the method decomposes the face as a linear combination of its neighbors from the known features from the face models. Then, the method determines weights for the known features to weight the coefficients of the known features. Particular embodiments use a non-linear weighting function to learn the weights that provides more accurate labels.