Dynamic Facial Expression Recognition Using Dempster-Shafer Theory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional facial expression recognition methods face challenges due to weak expressive features and sample imbalance in datasets, leading to poor performance in accurately classifying dynamic facial expressions.
Innovation Solution
A dynamic facial expression recognition method based on Dempster-Shafer theory, utilizing a classifier guided by the DS theory, a feature extraction module with an attention mechanism and multi-layer perceptron, and a spatio-temporal feature fusion module to improve recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional handmade features are used for expression classification, then the method is simple to implement, but the expressive force is weak and recognition accuracy is poor
Solution Approach 1:
The patent replaces traditional mechanical handcrafted feature extraction with a deep learning-based automatic feature extraction system. The neural network automatically learns hierarchical features from raw video frames, substituting the manual feature engineering process with an automated data-driven approach that achieves superior recognition accuracy without requiring domain expertise in feature design.
Solution Approach 2:
The patent transforms the feature representation parameters from fixed handcrafted descriptors to dynamic learned features through multiple neural network layers. By changing the parameter space from predefined features to learned embeddings, the system adapts features to the specific dataset and task, significantly improving expressive power and recognition performance.
2Measurement precision
If deep learning methods are used for FER, then feature extraction capability is enhanced, but sample imbalance in datasets leads to insufficient feature learning for minority classes
Solution Approach 1:
The patent applies partial action by focusing computational resources and sampling efforts on minority classes. Through oversampling minority class samples and adjusting sampling strategies, the system provides excessive representation to underrepresented classes during training, ensuring they receive sufficient attention for robust feature learning without compromising overall system performance.
Solution Approach 2:
The patent modifies training parameters including class weights, sampling probabilities, and loss function parameters to compensate for sample imbalance. By dynamically adjusting these parameters based on class distribution, the system ensures minority classes receive appropriate attention during gradient updates, improving their feature learning and recognition reliability.
3Measurement precision
If conventional classifiers are used, then the model structure is simple, but they cannot effectively handle sample imbalance and reduce recognition accuracy
Solution Approach 1:
The patent introduces an intermediary module between feature extraction and classification that handles sample imbalance through reweighting and resampling mechanisms. This intermediary layer processes imbalanced data before classification, acting as a mediator that prepares balanced representations for the classifier, thereby improving accuracy without requiring the classifier itself to be highly complex.
Solution Approach 2:
The patent employs dynamic classification strategies that adapt during training based on class distribution and performance metrics. The classifier structure and parameters are dynamically adjusted through techniques like focal loss and adaptive sampling, allowing the system to respond to sample imbalance in real-time during the learning process and maintain high accuracy across all classes.
Data Source
AI summary
A dynamic facial expression recognition (FER) method based on a Dempster-Shafer (DS) theory improves a feature extraction effect of an expression video through multi-feature fusion, and deeply learns an imbalanced dynamic expression feature by using the DS theory, multi-branch convolution, and an attention mechanism. Compared with other methods, the dynamic FER method scientifically and effectively reduces an impact of sample imbalance on expression recognition, fully utilizes a spatio-temporal feature to mine potential semantic information of the video expression to perform expression classification, thereby improving reliability and accuracy and meeting a demand for the expression recognition.