Facial Expression Recognition via Autoencoder Feature Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial expression recognition systems face difficulties in accurately identifying minute changes in facial expressions and often misclassify similar expressions, leading to inaccurate recognition.
Innovation Solution
The method involves generating a reference image from an input image using an auto encoder and extractor trained to decrease the difference between the input and target images, allowing for accurate recognition by comparing feature vectors extracted from both images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If direct pixel comparison is used for facial expression recognition, then the process is simple, but the recognition accuracy is low for minute changes
Solution Approach 1:
The patent introduces an autoencoder as an intermediary component that transforms input images into a compressed latent representation and then reconstructs them. This intermediary transformation enables the system to capture subtle facial expression changes that are not apparent in direct pixel comparison, thereby improving recognition accuracy while maintaining manageable system complexity through the use of pre-trained neural network models.
Solution Approach 2:
The patent replaces the traditional mechanical approach of direct pixel-by-pixel comparison with a neural network-based feature extraction and comparison system. The autoencoder-based approach substitutes simple arithmetic operations with learned feature representations, enabling the system to detect minute facial expression changes that would be imperceptible through direct pixel comparison.
2Measurement precision
If feature vectors are extracted and compared instead of direct image comparison, then recognition accuracy improves, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by using pre-trained autoencoder models that have already learned optimal feature extraction patterns during the training phase. When processing new images, the system only needs to perform forward propagation through the trained network, leveraging the pre-learned weights and biases to efficiently extract meaningful features without requiring intensive real-time computation, thus reducing the computational power needed at inference time.
3Measurement precision
If the autoencoder is trained to decrease difference between input and target images, then feature extraction accuracy improves, but training time increases
Solution Approach 1:
The patent employs parameter changes by adjusting the autoencoder's architecture parameters (such as latent space dimensionality, number of layers, and activation functions) and training parameters (learning rate, batch size, and loss function weights) to optimize the balance between feature extraction accuracy and training efficiency. By carefully tuning these parameters, the system achieves high recognition accuracy while reducing the computational burden and training time required.
Data Source
AI summary
A method and apparatus for recognizing a facial expression is provided. The method of recognizing the facial expression includes extracting a feature from an input image, extracting a feature from a reference image, and recognizing a user facial expression based on a difference between the feature of the input image and the feature of the reference image.