Social Relation Identification via Shared Facial Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods fail to accurately identify social relations between individuals in images, as existing datasets lack comprehensive facial attribute annotations and exhibit varying statistical distributions, making it difficult to learn rich representations for social relation inference.
Innovation Solution
A system utilizing a convolutional neural network with a training process that adjusts weights to extract shared facial features from spatial cues, including face positions and relative scales, to predict social relations by bridging heterogeneous datasets with weak constraints derived from face part appearances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing datasets are used for training, then the system can perform social relation identification, but the datasets lack comprehensive facial attribute annotations and exhibit varying statistical distributions, reducing identification accuracy
Solution Approach 1:
The patent introduces a weak constraint layer as an intermediary between heterogeneous datasets and the social relation identification task. This layer extracts facial part appearances (eyes, mouth, nose) from diverse datasets and uses them as bridge features to train the convolutional neural network, enabling the system to handle varying statistical distributions and missing attribute labels while maintaining identification accuracy
Solution Approach 2:
The patent segments the face into distinct parts (eyes, mouth, nose) and extracts features from each segment independently. This segmentation allows the system to handle incomplete annotations by focusing on available facial part information, thereby improving robustness across heterogeneous datasets with varying annotation completeness
2Adaptability or versatility
If heterogeneous datasets with varying statistical distributions are used, then the system can learn from diverse data sources, but the varying distributions make it difficult to learn rich representations for social relation inference
Solution Approach 1:
The patent changes the parameter space by transforming diverse facial attributes into a unified representation based on facial part appearances. By focusing on common facial parts (eyes, mouth, nose) that appear across different datasets, the system normalizes the statistical distributions and learns consistent feature representations that generalize across heterogeneous data sources
Solution Approach 2:
The patent creates a universal feature extraction mechanism that works across multiple datasets with different statistical distributions. The convolutional neural network, trained on facial part appearances from diverse datasets, becomes a multi-functional system that can handle various data sources while maintaining consistent and accurate social relation inference capabilities
3Measurement precision
If comprehensive facial attribute annotations are required, then the system can achieve high identification accuracy, but the annotations are missing in existing datasets
Solution Approach 1:
The patent extracts useful information from available facial part appearances rather than requiring complete attribute annotations. By taking out and focusing on visible facial parts (eyes, mouth, nose), the system can infer social relations even when comprehensive attribute labels are missing, thereby reducing information loss and maintaining identification accuracy
Data Source
AI summary
A method for identifying social relation of persons in an image, including: generating face regions for faces of the persons in the image; determining at least one spatial cue for each of the faces; extracting features related to social relation for each face from the face regions; determining a shared facial feature from the extracted features and the determined spatial cue, the determined feature being shared by multiple the social relation inferences; and predicting the social relation of the persons from the shared facial feature.