Multi-Domain Adversarial Learning for Facial Expression Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning methods for compound facial expression recognition require a large number of training samples and result in excessive network parameters, making them inefficient for scenarios with few data samples and low precision.
Innovation Solution
A multi-domain adversarial learning method that divides datasets into face, contour-independent, and contour-related sub-domains, using generators and discriminators to fuse features and reduce network parameters while maintaining high recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If general convolutional neural network methods (MobileNet, VGG, Inception-ResNet) are used for compound expression recognition, then recognition capability is achieved, but a large number of training samples are required and network parameters become excessive
Solution Approach 1:
The patent segments the compound expression recognition task into multiple domains: basic expression domain, compound expression domain, and identity domain. Each domain is processed by specialized network branches that extract features independently, then fuse them for final recognition. This segmentation allows the model to learn from fewer samples by focusing on domain-specific patterns rather than requiring comprehensive general data.
Solution Approach 2:
The patent introduces domain adaptation by adding a domain classification dimension that distinguishes between basic expressions, compound expressions, and identities. This dimensional extension allows the model to leverage transfer learning across domains, improving recognition accuracy with fewer compound expression samples by utilizing knowledge from basic expression and identity domains.
2Adaptability or versatility
If multiple residual blocks are used to detect both compound and basic expressions, then comprehensive expression detection is achieved, but network parameters become excessive and efficiency decreases
Solution Approach 1:
The patent divides the expression detection task into separate branches: one for basic expressions and another for compound expressions. Each branch uses optimized residual blocks tailored to its specific detection goals, avoiding the redundancy of using 15 residual blocks for both tasks simultaneously. This segmented approach maintains comprehensive detection capability while significantly reducing total network parameters.
Solution Approach 2:
The patent designs a multi-functional network architecture where shared backbone features serve multiple detection purposes. The base network extracts general facial features that are then routed to specialized detection heads for basic expressions, compound expressions, and identities, allowing one network structure to perform multiple functions efficiently without requiring separate full-scale networks for each task.
3Measurement precision
If two-stage fine-grained facial expression recognition (MSAU-Net) is used, then detailed AU feature detection is achieved, but parameter scale becomes excessively large during inference
Solution Approach 1:
The patent extracts and focuses specifically on Action Unit (AU) related features that are critical for compound expression recognition, rather than processing all facial features equally. By identifying and extracting only the relevant AU regions and features needed for compound expression detection, the model achieves detailed feature precision while reducing the overall parameter scale by eliminating unnecessary processing of non-critical facial regions.
Solution Approach 2:
The patent applies local quality enhancement by using specialized processing for different facial regions. Instead of uniform two-stage processing across the entire face, the model applies enhanced feature extraction specifically to AU-related local regions while using standard processing elsewhere, achieving detailed AU detection precision with reduced computational overhead and parameter scale.
Data Source
AI summary
Disclosed is a compound expression recognition method with few samples of multi-domain adversarial learning. To extract compound expression features with diversity and complexity with few samples, multiple small sample datasets are fused, and divided into expression sub-domains, and multi-domain adversarial learning is performed to improve the performance of compound expression recognition. Based on the generative adversarial network framework, the face domain and the contour-independent compound expression domain are fused in the generative network to enhance diversity and complexity, and two discriminators are designed to guide the generator. The face discriminator uses the face domain to guide the generator and identify the generator to generate expression-independent face identity attributes, so that the generator has identity diversity. The compound expression fusing discriminator fuses the basic expression domain and the contour-related compound expression domain together to guide the generator and identify the complexity of the expressions generated by the generator.