Facial Expression Recognition via Key-Point Cascaded Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current expression recognition methods using deep learning are complex and involve many parameters, making it difficult to achieve accurate classification of facial expressions due to the complexity of human facial expressions.
Innovation Solution
The proposed method involves face key-point position detection, followed by sequential feature processing through four cascaded convolutional modules, global average pooling, key-point feature extraction, and connection of feature vectors to a fully connected layer for classification by a trained neural network classifier.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face images are extracted using face key-points and enlarged to face image size for input to deep learning network, then expression recognition can be performed, but the model structure becomes complex and involves more parameters
Solution Approach 1:
The patent extracts only the essential feature information from face images through key-point detection and selective feature map processing, rather than using complete enlarged face images. This extraction approach reduces the data volume and model complexity while retaining the critical expression features needed for accurate recognition.
Solution Approach 2:
The patent divides the feature processing into multiple stages using cascaded convolutional modules, where each module processes feature maps at different levels. This segmentation allows the model to progressively extract expression features without requiring a single complex processing stage, thereby reducing overall model complexity while maintaining accuracy.
2Reliability
If complete face images are input to deep learning network for expression recognition, then comprehensive feature analysis can be performed, but the number of parameters increases significantly
Solution Approach 1:
The patent extracts essential expression features through key-point detection and selective processing of feature response maps from cascaded convolutional modules. This extraction eliminates redundant information from complete face images, significantly reducing the number of parameters required while preserving the reliability of expression classification through focused feature analysis.
Solution Approach 2:
Instead of processing complete face images, the patent applies partial processing by focusing on specific feature response maps from intermediate convolutional modules that contain the most discriminative expression information. This partial action approach reduces parameter quantity while maintaining sufficient reliability for accurate expression recognition.
Data Source
AI summary
An expression recognition method and apparatus, a computer device, and a readable storage medium are provided. The method includes: performing face key-point position detection on a face image to obtain face key-point position information; and obtaining expression class information of the face image using four cascaded convolutional modules and a trained neural network classifier according to the face image and the face key-point position information.